diff --git a/modules/core/CMakeLists.txt b/modules/core/CMakeLists.txt
index 0cf404e805..1895333fed 100644
--- a/modules/core/CMakeLists.txt
+++ b/modules/core/CMakeLists.txt
@@ -8,6 +8,7 @@ ocv_add_dispatched_file(convert_scale SSE2 AVX2)
 ocv_add_dispatched_file(count_non_zero SSE2 AVX2)
 ocv_add_dispatched_file(matmul SSE2 AVX2)
 ocv_add_dispatched_file(mean SSE2 AVX2)
+ocv_add_dispatched_file(split SSE2 AVX2)
 ocv_add_dispatched_file(sum SSE2 AVX2)
 
 # dispatching for accuracy tests
diff --git a/modules/core/src/split.cpp b/modules/core/src/split.dispatch.cpp
similarity index 56%
rename from modules/core/src/split.cpp
rename to modules/core/src/split.dispatch.cpp
index ac1b7217bb..e41eaf1fe7 100644
--- a/modules/core/src/split.cpp
+++ b/modules/core/src/split.dispatch.cpp
@@ -6,213 +6,44 @@
 #include "precomp.hpp"
 #include "opencl_kernels_core.hpp"
 
+#include "split.simd.hpp"
+#include "split.simd_declarations.hpp" // defines CV_CPU_DISPATCH_MODES_ALL=AVX2,...,BASELINE based on CMakeLists.txt content
+
 namespace cv { namespace hal {
 
-#if CV_SIMD
-// see the comments for vecmerge_ in merge.cpp
-template<typename T, typename VecT> static void
-vecsplit_( const T* src, T** dst, int len, int cn )
-{
-    const int VECSZ = VecT::nlanes;
-    int i, i0 = 0;
-    T* dst0 = dst[0];
-    T* dst1 = dst[1];
-
-    int r0 = (int)((size_t)(void*)dst0 % (VECSZ*sizeof(T)));
-    int r1 = (int)((size_t)(void*)dst1 % (VECSZ*sizeof(T)));
-    int r2 = cn > 2 ? (int)((size_t)(void*)dst[2] % (VECSZ*sizeof(T))) : r0;
-    int r3 = cn > 3 ? (int)((size_t)(void*)dst[3] % (VECSZ*sizeof(T))) : r0;
-
-    hal::StoreMode mode = hal::STORE_ALIGNED_NOCACHE;
-    if( (r0|r1|r2|r3) != 0 )
-    {
-        mode = hal::STORE_UNALIGNED;
-        if( r0 == r1 && r0 == r2 && r0 == r3 && r0 % sizeof(T) == 0 && len > VECSZ*2 )
-            i0 = VECSZ - (r0 / sizeof(T));
-    }
-
-    if( cn == 2 )
-    {
-        for( i = 0; i < len; i += VECSZ )
-        {
-            if( i > len - VECSZ )
-            {
-                i = len - VECSZ;
-                mode = hal::STORE_UNALIGNED;
-            }
-            VecT a, b;
-            v_load_deinterleave(src + i*cn, a, b);
-            v_store(dst0 + i, a, mode);
-            v_store(dst1 + i, b, mode);
-            if( i < i0 )
-            {
-                i = i0 - VECSZ;
-                mode = hal::STORE_ALIGNED_NOCACHE;
-            }
-        }
-    }
-    else if( cn == 3 )
-    {
-        T* dst2 = dst[2];
-        for( i = 0; i < len; i += VECSZ )
-        {
-            if( i > len - VECSZ )
-            {
-                i = len - VECSZ;
-                mode = hal::STORE_UNALIGNED;
-            }
-            VecT a, b, c;
-            v_load_deinterleave(src + i*cn, a, b, c);
-            v_store(dst0 + i, a, mode);
-            v_store(dst1 + i, b, mode);
-            v_store(dst2 + i, c, mode);
-            if( i < i0 )
-            {
-                i = i0 - VECSZ;
-                mode = hal::STORE_ALIGNED_NOCACHE;
-            }
-        }
-    }
-    else
-    {
-        CV_Assert( cn == 4 );
-        T* dst2 = dst[2];
-        T* dst3 = dst[3];
-        for( i = 0; i < len; i += VECSZ )
-        {
-            if( i > len - VECSZ )
-            {
-                i = len - VECSZ;
-                mode = hal::STORE_UNALIGNED;
-            }
-            VecT a, b, c, d;
-            v_load_deinterleave(src + i*cn, a, b, c, d);
-            v_store(dst0 + i, a, mode);
-            v_store(dst1 + i, b, mode);
-            v_store(dst2 + i, c, mode);
-            v_store(dst3 + i, d, mode);
-            if( i < i0 )
-            {
-                i = i0 - VECSZ;
-                mode = hal::STORE_ALIGNED_NOCACHE;
-            }
-        }
-    }
-    vx_cleanup();
-}
-#endif
-
-template<typename T> static void
-split_( const T* src, T** dst, int len, int cn )
-{
-    int k = cn % 4 ? cn % 4 : 4;
-    int i, j;
-    if( k == 1 )
-    {
-        T* dst0 = dst[0];
-
-        if(cn == 1)
-        {
-            memcpy(dst0, src, len * sizeof(T));
-        }
-        else
-        {
-            for( i = 0, j = 0 ; i < len; i++, j += cn )
-                dst0[i] = src[j];
-        }
-    }
-    else if( k == 2 )
-    {
-        T *dst0 = dst[0], *dst1 = dst[1];
-        i = j = 0;
-
-        for( ; i < len; i++, j += cn )
-        {
-            dst0[i] = src[j];
-            dst1[i] = src[j+1];
-        }
-    }
-    else if( k == 3 )
-    {
-        T *dst0 = dst[0], *dst1 = dst[1], *dst2 = dst[2];
-        i = j = 0;
-
-        for( ; i < len; i++, j += cn )
-        {
-            dst0[i] = src[j];
-            dst1[i] = src[j+1];
-            dst2[i] = src[j+2];
-        }
-    }
-    else
-    {
-        T *dst0 = dst[0], *dst1 = dst[1], *dst2 = dst[2], *dst3 = dst[3];
-        i = j = 0;
-
-        for( ; i < len; i++, j += cn )
-        {
-            dst0[i] = src[j]; dst1[i] = src[j+1];
-            dst2[i] = src[j+2]; dst3[i] = src[j+3];
-        }
-    }
-
-    for( ; k < cn; k += 4 )
-    {
-        T *dst0 = dst[k], *dst1 = dst[k+1], *dst2 = dst[k+2], *dst3 = dst[k+3];
-        for( i = 0, j = k; i < len; i++, j += cn )
-        {
-            dst0[i] = src[j]; dst1[i] = src[j+1];
-            dst2[i] = src[j+2]; dst3[i] = src[j+3];
-        }
-    }
-}
-
 void split8u(const uchar* src, uchar** dst, int len, int cn )
 {
+    CV_INSTRUMENT_REGION();
     CALL_HAL(split8u, cv_hal_split8u, src,dst, len, cn)
-
-#if CV_SIMD
-    if( len >= v_uint8::nlanes && 2 <= cn && cn <= 4 )
-        vecsplit_<uchar, v_uint8>(src, dst, len, cn);
-    else
-#endif
-        split_(src, dst, len, cn);
+    CV_CPU_DISPATCH(split8u, (src, dst, len, cn),
+        CV_CPU_DISPATCH_MODES_ALL);
 }
 
 void split16u(const ushort* src, ushort** dst, int len, int cn )
 {
+    CV_INSTRUMENT_REGION();
     CALL_HAL(split16u, cv_hal_split16u, src,dst, len, cn)
-#if CV_SIMD
-    if( len >= v_uint16::nlanes && 2 <= cn && cn <= 4 )
-        vecsplit_<ushort, v_uint16>(src, dst, len, cn);
-    else
-#endif
-        split_(src, dst, len, cn);
+    CV_CPU_DISPATCH(split16u, (src, dst, len, cn),
+        CV_CPU_DISPATCH_MODES_ALL);
 }
 
 void split32s(const int* src, int** dst, int len, int cn )
 {
+    CV_INSTRUMENT_REGION();
     CALL_HAL(split32s, cv_hal_split32s, src,dst, len, cn)
-#if CV_SIMD
-    if( len >= v_uint32::nlanes && 2 <= cn && cn <= 4 )
-        vecsplit_<int, v_int32>(src, dst, len, cn);
-    else
-#endif
-        split_(src, dst, len, cn);
+    CV_CPU_DISPATCH(split32s, (src, dst, len, cn),
+        CV_CPU_DISPATCH_MODES_ALL);
 }
 
 void split64s(const int64* src, int64** dst, int len, int cn )
 {
+    CV_INSTRUMENT_REGION();
     CALL_HAL(split64s, cv_hal_split64s, src,dst, len, cn)
-#if CV_SIMD
-    if( len >= v_int64::nlanes && 2 <= cn && cn <= 4 )
-        vecsplit_<int64, v_int64>(src, dst, len, cn);
-    else
-#endif
-        split_(src, dst, len, cn);
+    CV_CPU_DISPATCH(split64s, (src, dst, len, cn),
+        CV_CPU_DISPATCH_MODES_ALL);
 }
 
-}} // cv::hal::
+} // namespace cv::hal::
 
 /****************************************************************************************\
 *                                       split & merge                                    *
@@ -233,7 +64,6 @@ static SplitFunc getSplitFunc(int depth)
 
 #ifdef HAVE_IPP
 
-namespace cv {
 static bool ipp_split(const Mat& src, Mat* mv, int channels)
 {
 #ifdef HAVE_IPP_IW_LL
@@ -282,10 +112,9 @@ static bool ipp_split(const Mat& src, Mat* mv, int channels)
     return false;
 #endif
 }
-}
 #endif
 
-void cv::split(const Mat& src, Mat* mv)
+void split(const Mat& src, Mat* mv)
 {
     CV_INSTRUMENT_REGION();
 
@@ -341,8 +170,6 @@ void cv::split(const Mat& src, Mat* mv)
 
 #ifdef HAVE_OPENCL
 
-namespace cv {
-
 static bool ocl_split( InputArray _m, OutputArrayOfArrays _mv )
 {
     int type = _m.type(), depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type),
@@ -381,11 +208,9 @@ static bool ocl_split( InputArray _m, OutputArrayOfArrays _mv )
     return k.run(2, globalsize, NULL, false);
 }
 
-}
-
 #endif
 
-void cv::split(InputArray _m, OutputArrayOfArrays _mv)
+void split(InputArray _m, OutputArrayOfArrays _mv)
 {
     CV_INSTRUMENT_REGION();
 
@@ -411,3 +236,5 @@ void cv::split(InputArray _m, OutputArrayOfArrays _mv)
 
     split(m, &dst[0]);
 }
+
+} // namespace
diff --git a/modules/core/src/split.simd.hpp b/modules/core/src/split.simd.hpp
new file mode 100644
index 0000000000..25e90c0063
--- /dev/null
+++ b/modules/core/src/split.simd.hpp
@@ -0,0 +1,223 @@
+// This file is part of OpenCV project.
+// It is subject to the license terms in the LICENSE file found in the top-level directory
+// of this distribution and at http://opencv.org/license.html
+
+
+#include "precomp.hpp"
+
+namespace cv { namespace hal {
+CV_CPU_OPTIMIZATION_NAMESPACE_BEGIN
+
+void split8u(const uchar* src, uchar** dst, int len, int cn);
+void split16u(const ushort* src, ushort** dst, int len, int cn);
+void split32s(const int* src, int** dst, int len, int cn);
+void split64s(const int64* src, int64** dst, int len, int cn);
+
+#ifndef CV_CPU_OPTIMIZATION_DECLARATIONS_ONLY
+
+#if CV_SIMD
+// see the comments for vecmerge_ in merge.cpp
+template<typename T, typename VecT> static void
+vecsplit_( const T* src, T** dst, int len, int cn )
+{
+    const int VECSZ = VecT::nlanes;
+    int i, i0 = 0;
+    T* dst0 = dst[0];
+    T* dst1 = dst[1];
+
+    int r0 = (int)((size_t)(void*)dst0 % (VECSZ*sizeof(T)));
+    int r1 = (int)((size_t)(void*)dst1 % (VECSZ*sizeof(T)));
+    int r2 = cn > 2 ? (int)((size_t)(void*)dst[2] % (VECSZ*sizeof(T))) : r0;
+    int r3 = cn > 3 ? (int)((size_t)(void*)dst[3] % (VECSZ*sizeof(T))) : r0;
+
+    hal::StoreMode mode = hal::STORE_ALIGNED_NOCACHE;
+    if( (r0|r1|r2|r3) != 0 )
+    {
+        mode = hal::STORE_UNALIGNED;
+        if( r0 == r1 && r0 == r2 && r0 == r3 && r0 % sizeof(T) == 0 && len > VECSZ*2 )
+            i0 = VECSZ - (r0 / sizeof(T));
+    }
+
+    if( cn == 2 )
+    {
+        for( i = 0; i < len; i += VECSZ )
+        {
+            if( i > len - VECSZ )
+            {
+                i = len - VECSZ;
+                mode = hal::STORE_UNALIGNED;
+            }
+            VecT a, b;
+            v_load_deinterleave(src + i*cn, a, b);
+            v_store(dst0 + i, a, mode);
+            v_store(dst1 + i, b, mode);
+            if( i < i0 )
+            {
+                i = i0 - VECSZ;
+                mode = hal::STORE_ALIGNED_NOCACHE;
+            }
+        }
+    }
+    else if( cn == 3 )
+    {
+        T* dst2 = dst[2];
+        for( i = 0; i < len; i += VECSZ )
+        {
+            if( i > len - VECSZ )
+            {
+                i = len - VECSZ;
+                mode = hal::STORE_UNALIGNED;
+            }
+            VecT a, b, c;
+            v_load_deinterleave(src + i*cn, a, b, c);
+            v_store(dst0 + i, a, mode);
+            v_store(dst1 + i, b, mode);
+            v_store(dst2 + i, c, mode);
+            if( i < i0 )
+            {
+                i = i0 - VECSZ;
+                mode = hal::STORE_ALIGNED_NOCACHE;
+            }
+        }
+    }
+    else
+    {
+        CV_Assert( cn == 4 );
+        T* dst2 = dst[2];
+        T* dst3 = dst[3];
+        for( i = 0; i < len; i += VECSZ )
+        {
+            if( i > len - VECSZ )
+            {
+                i = len - VECSZ;
+                mode = hal::STORE_UNALIGNED;
+            }
+            VecT a, b, c, d;
+            v_load_deinterleave(src + i*cn, a, b, c, d);
+            v_store(dst0 + i, a, mode);
+            v_store(dst1 + i, b, mode);
+            v_store(dst2 + i, c, mode);
+            v_store(dst3 + i, d, mode);
+            if( i < i0 )
+            {
+                i = i0 - VECSZ;
+                mode = hal::STORE_ALIGNED_NOCACHE;
+            }
+        }
+    }
+    vx_cleanup();
+}
+#endif
+
+template<typename T> static void
+split_( const T* src, T** dst, int len, int cn )
+{
+    int k = cn % 4 ? cn % 4 : 4;
+    int i, j;
+    if( k == 1 )
+    {
+        T* dst0 = dst[0];
+
+        if(cn == 1)
+        {
+            memcpy(dst0, src, len * sizeof(T));
+        }
+        else
+        {
+            for( i = 0, j = 0 ; i < len; i++, j += cn )
+                dst0[i] = src[j];
+        }
+    }
+    else if( k == 2 )
+    {
+        T *dst0 = dst[0], *dst1 = dst[1];
+        i = j = 0;
+
+        for( ; i < len; i++, j += cn )
+        {
+            dst0[i] = src[j];
+            dst1[i] = src[j+1];
+        }
+    }
+    else if( k == 3 )
+    {
+        T *dst0 = dst[0], *dst1 = dst[1], *dst2 = dst[2];
+        i = j = 0;
+
+        for( ; i < len; i++, j += cn )
+        {
+            dst0[i] = src[j];
+            dst1[i] = src[j+1];
+            dst2[i] = src[j+2];
+        }
+    }
+    else
+    {
+        T *dst0 = dst[0], *dst1 = dst[1], *dst2 = dst[2], *dst3 = dst[3];
+        i = j = 0;
+
+        for( ; i < len; i++, j += cn )
+        {
+            dst0[i] = src[j]; dst1[i] = src[j+1];
+            dst2[i] = src[j+2]; dst3[i] = src[j+3];
+        }
+    }
+
+    for( ; k < cn; k += 4 )
+    {
+        T *dst0 = dst[k], *dst1 = dst[k+1], *dst2 = dst[k+2], *dst3 = dst[k+3];
+        for( i = 0, j = k; i < len; i++, j += cn )
+        {
+            dst0[i] = src[j]; dst1[i] = src[j+1];
+            dst2[i] = src[j+2]; dst3[i] = src[j+3];
+        }
+    }
+}
+
+void split8u(const uchar* src, uchar** dst, int len, int cn )
+{
+    CV_INSTRUMENT_REGION();
+#if CV_SIMD
+    if( len >= v_uint8::nlanes && 2 <= cn && cn <= 4 )
+        vecsplit_<uchar, v_uint8>(src, dst, len, cn);
+    else
+#endif
+        split_(src, dst, len, cn);
+}
+
+void split16u(const ushort* src, ushort** dst, int len, int cn )
+{
+    CV_INSTRUMENT_REGION();
+#if CV_SIMD
+    if( len >= v_uint16::nlanes && 2 <= cn && cn <= 4 )
+        vecsplit_<ushort, v_uint16>(src, dst, len, cn);
+    else
+#endif
+        split_(src, dst, len, cn);
+}
+
+void split32s(const int* src, int** dst, int len, int cn )
+{
+    CV_INSTRUMENT_REGION();
+#if CV_SIMD
+    if( len >= v_uint32::nlanes && 2 <= cn && cn <= 4 )
+        vecsplit_<int, v_int32>(src, dst, len, cn);
+    else
+#endif
+        split_(src, dst, len, cn);
+}
+
+void split64s(const int64* src, int64** dst, int len, int cn )
+{
+    CV_INSTRUMENT_REGION();
+#if CV_SIMD
+    if( len >= v_int64::nlanes && 2 <= cn && cn <= 4 )
+        vecsplit_<int64, v_int64>(src, dst, len, cn);
+    else
+#endif
+        split_(src, dst, len, cn);
+}
+
+#endif
+CV_CPU_OPTIMIZATION_NAMESPACE_END
+}} // namespace
\ No newline at end of file