diff --git a/src/core/CL/CLKernels.h b/src/core/CL/CLKernels.h
index 09f8109..d2d5b92 100644
--- a/src/core/CL/CLKernels.h
+++ b/src/core/CL/CLKernels.h
@@ -31,7 +31,6 @@
 #include "src/core/CL/kernels/CLBitwiseKernel.h"
 #include "src/core/CL/kernels/CLBoundingBoxTransformKernel.h"
 #include "src/core/CL/kernels/CLChannelShuffleLayerKernel.h"
-#include "src/core/CL/kernels/CLCol2ImKernel.h"
 #include "src/core/CL/kernels/CLComparisonKernel.h"
 #include "src/core/CL/kernels/CLDeconvolutionLayerUpsampleKernel.h"
 #include "src/core/CL/kernels/CLDeconvolutionReshapeOutputKernel.h"
diff --git a/src/core/CL/kernels/CLCol2ImKernel.cpp b/src/core/CL/kernels/CLCol2ImKernel.cpp
deleted file mode 100644
index 6d7b834..0000000
--- a/src/core/CL/kernels/CLCol2ImKernel.cpp
+++ /dev/null
@@ -1,173 +0,0 @@
-/*
- * Copyright (c) 2017-2021 Arm Limited.
- *
- * SPDX-License-Identifier: MIT
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-#include "src/core/CL/kernels/CLCol2ImKernel.h"
-
-#include "arm_compute/core/CL/CLHelpers.h"
-#include "arm_compute/core/CL/CLKernelLibrary.h"
-#include "arm_compute/core/CL/ICLTensor.h"
-#include "arm_compute/core/CL/OpenCL.h"
-#include "arm_compute/core/Helpers.h"
-#include "arm_compute/core/utils/misc/ShapeCalculator.h"
-#include "src/core/CL/CLValidate.h"
-#include "src/core/helpers/AutoConfiguration.h"
-#include "src/core/helpers/WindowHelpers.h"
-#include "support/StringSupport.h"
-
-#include <cmath>
-
-using namespace arm_compute::misc::shape_calculator;
-
-namespace arm_compute
-{
-namespace
-{
-Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const Size2D &convolved_dims, unsigned int num_groups)
-{
-    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
-    ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(input);
-    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::F16, DataType::F32);
-
-    // Checks performed when output is configured
-    if(output->total_size() != 0)
-    {
-        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(output->tensor_shape(), compute_col2im_shape(*input, convolved_dims, true, num_groups));
-        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
-        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(input, output);
-        ARM_COMPUTE_RETURN_ERROR_ON_MSG(output->data_layout() != DataLayout::NCHW, "Col2Im output's data layout must always be NCHW");
-    }
-
-    return Status{};
-}
-
-std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *output, const Size2D &convolved_dims, unsigned int num_groups)
-{
-    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
-    // Output auto inizialitation if not yet initialized
-    auto_init_if_empty(*output, input->clone()->set_tensor_shape(compute_col2im_shape(*input, convolved_dims, true, num_groups)).set_data_layout(DataLayout::NCHW));
-
-    constexpr unsigned int num_elems_read_per_iteration = 8;
-
-    // Configure window
-    Window win = calculate_max_window(*input, Steps(num_elems_read_per_iteration));
-
-    // Update window and padding just for the input tensor as we cannot access out-of-bounds elements in the output one
-    AccessWindowHorizontal input_access(input, 0, num_elems_read_per_iteration);
-    bool                   window_changed = update_window_and_padding(win, input_access);
-
-    Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
-    return std::make_pair(err, win);
-}
-} // namespace
-
-CLCol2ImKernel::CLCol2ImKernel()
-    : _input(nullptr), _output(nullptr), _convolved_dims()
-{
-    _type = CLKernelType::ELEMENTWISE;
-}
-
-void CLCol2ImKernel::configure(const ICLTensor *input, ICLTensor *output, const Size2D &convolved_dims, unsigned int num_groups)
-{
-    configure(CLKernelLibrary::get().get_compile_context(), input, output, convolved_dims, num_groups);
-}
-
-void CLCol2ImKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, const Size2D &convolved_dims, unsigned int num_groups)
-{
-    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
-
-    // Perform validation step
-    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), convolved_dims, num_groups));
-
-    _input          = input;
-    _output         = output;
-    _convolved_dims = convolved_dims;
-
-    const DataType data_type = input->info()->data_type();
-
-    // Create kernel
-    CLBuildOptions build_opts;
-    build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(data_type));
-    build_opts.add_option("-DELEMENT_SIZE=" + support::cpp11::to_string(input->info()->element_size()));
-    build_opts.add_option("-DWIDTH_INPUT=" + support::cpp11::to_string(input->info()->dimension(0)));
-    build_opts.add_option("-DWIDTH_OUTPUT=" + support::cpp11::to_string(_convolved_dims.width));
-    build_opts.add_option("-DNUM_GROUPS=" + support::cpp11::to_string(num_groups));
-
-    _kernel = create_kernel(compile_context, "col2im", build_opts.options());
-
-    // Configure kernel window
-    auto win_config = validate_and_configure_window(input->info(), output->info(), _convolved_dims, num_groups);
-    ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
-    ICLKernel::configure_internal(win_config.second);
-
-    // Set config_id for enabling LWS tuning
-    _config_id = "col2im_";
-    _config_id += lower_string(string_from_data_type(input->info()->data_type()));
-    _config_id += "_";
-    _config_id += support::cpp11::to_string(num_groups);
-    _config_id += "_";
-    _config_id += support::cpp11::to_string(input->info()->dimension(0));
-    _config_id += "_";
-    _config_id += support::cpp11::to_string(input->info()->dimension(1));
-    _config_id += "_";
-    _config_id += support::cpp11::to_string(output->info()->dimension(0));
-    _config_id += "_";
-    _config_id += support::cpp11::to_string(output->info()->dimension(1));
-}
-
-Status CLCol2ImKernel::validate(const ITensorInfo *input, const ITensorInfo *output, const Size2D &convolved_dims, unsigned int num_groups)
-{
-    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, convolved_dims, num_groups));
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), output->clone().get(), convolved_dims, num_groups).first);
-    return Status{};
-}
-
-void CLCol2ImKernel::run(const Window &window, cl::CommandQueue &queue)
-{
-    ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
-    ARM_COMPUTE_ERROR_ON_MISMATCHING_WINDOWS(ICLKernel::window(), window);
-
-    bool is_collapsed     = false;
-    bool is_collapsed_out = false;
-
-    Window out_window;
-    out_window.use_tensor_dimensions(_output->info()->tensor_shape());
-
-    Window collapsed     = window.collapse_if_possible(ICLKernel::window(), Window::DimZ, &is_collapsed);
-    Window collapsed_out = out_window.collapse_if_possible(out_window, 3, &is_collapsed_out);
-
-    ARM_COMPUTE_ERROR_ON(is_collapsed != is_collapsed_out);
-
-    Window slice     = collapsed.first_slice_window_3D();
-    Window slice_out = collapsed_out.first_slice_window_4D();
-    do
-    {
-        // Set inputs
-        unsigned int idx = 0;
-        add_3D_tensor_argument(idx, _input, slice);
-        add_4D_tensor_argument(idx, _output, slice_out);
-        enqueue(queue, *this, slice, lws_hint());
-    }
-    while(collapsed.slide_window_slice_3D(slice) && collapsed_out.slide_window_slice_4D(slice_out));
-}
-} // namespace arm_compute
diff --git a/src/core/CL/kernels/CLCol2ImKernel.h b/src/core/CL/kernels/CLCol2ImKernel.h
deleted file mode 100644
index 710e048..0000000
--- a/src/core/CL/kernels/CLCol2ImKernel.h
+++ /dev/null
@@ -1,106 +0,0 @@
-/*
- * Copyright (c) 2017-2020 Arm Limited.
- *
- * SPDX-License-Identifier: MIT
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-#ifndef ARM_COMPUTE_CLCOL2IMKERNEL_H
-#define ARM_COMPUTE_CLCOL2IMKERNEL_H
-
-#include "src/core/CL/ICLKernel.h"
-
-namespace arm_compute
-{
-class ICLTensor;
-
-/** Interface for the col2im reshaping kernel.
- *
- * Rearranges each matrix column into image blocks. It's the inverse operation of @ref CLIm2ColKernel.
- *
- * For example, a vector of 9 elements can be reshaped to a block(image) of 3x3:
- *
- * @f[
- * \left( \begin{array}{ccccccccc}
- * a0 & a1 & a2 & a3 & a4 & a5 & a6 & a7 & a8 \\
- * \end{array} \right)
- * \rightarrow
- * \left( \begin{array}{ccc}
- * a0 & a1 & a2 \\
- * a3 & a4 & a5 \\
- * a6 & a7 & a8 \\
- * \end{array} \right)
- * @f]
- */
-class CLCol2ImKernel : public ICLKernel
-{
-public:
-    /** Default constructor */
-    CLCol2ImKernel();
-    /** Prevent instances of this class from being copied (As this class contains pointers) */
-    CLCol2ImKernel(const CLCol2ImKernel &) = delete;
-    /** Prevent instances of this class from being copied (As this class contains pointers) */
-    CLCol2ImKernel &operator=(const CLCol2ImKernel &) = delete;
-    /** Allow instances of this class to be moved */
-    CLCol2ImKernel(CLCol2ImKernel &&) = default;
-    /** Allow instances of this class to be moved */
-    CLCol2ImKernel &operator=(CLCol2ImKernel &&) = default;
-    /** Default destructor */
-    ~CLCol2ImKernel() = default;
-    /** Set the input and output of the kernel.
-     *
-     * @param[in]  input          The input tensor to convert. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32
-     * @param[out] output         The output tensor. 3 lower dimensions represent a single output [width, height, OFM],
-     *                            while the rest represent batch of outputs. Data types supported: Same as @p input. Data layout: NCHW
-     * @param[in]  convolved_dims Output convolved dimensions.
-     * @param[in]  num_groups     (Optional) Number of groups when performing a grouped convolution
-     */
-    void configure(const ICLTensor *input, ICLTensor *output, const Size2D &convolved_dims, unsigned int num_groups = 1);
-    /** Set the input and output of the kernel.
-     *
-     * @param[in]  compile_context The compile context to be used.
-     * @param[in]  input           The input tensor to convert. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32
-     * @param[out] output          The output tensor. 3 lower dimensions represent a single output [width, height, OFM],
-     *                             while the rest represent batch of outputs. Data types supported: Same as @p input. Data layout: NCHW
-     * @param[in]  convolved_dims  Output convolved dimensions.
-     * @param[in]  num_groups      (Optional) Number of groups when performing a grouped convolution
-     */
-    void configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, const Size2D &convolved_dims, unsigned int num_groups = 1);
-    /** Static function to check if given info will lead to a valid configuration of @ref CLCol2ImKernel
-     *
-     * @param[in] input          The input tensor to convert. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32
-     * @param[in] output         The output tensor. 3 lower dimensions represent a single output [width, height, OFM],
-     *                           while the rest represent batch of outputs. Data types supported: Same as @p input. Data layout: NCHW
-     * @param[in] convolved_dims Output convolved dimensions.
-     * @param[in] num_groups     (Optional) Number of groups when performing a grouped convolution
-     *
-     * @return a status
-     */
-    static Status validate(const ITensorInfo *input, const ITensorInfo *output, const Size2D &convolved_dims, unsigned int num_groups = 1);
-
-    // Inherited methods overridden:
-    void run(const Window &window, cl::CommandQueue &queue) override;
-
-public:
-    const ICLTensor *_input;
-    ICLTensor       *_output;
-    Size2D           _convolved_dims;
-};
-} // namespace arm_compute
-#endif /*ARM_COMPUTE_CLCOL2IMKERNEL_H */
diff --git a/src/core/gpu/cl/kernels/ClCol2ImKernel.cpp b/src/core/gpu/cl/kernels/ClCol2ImKernel.cpp
new file mode 100644
index 0000000..a3d5711
--- /dev/null
+++ b/src/core/gpu/cl/kernels/ClCol2ImKernel.cpp
@@ -0,0 +1,175 @@
+/*
+ * Copyright (c) 2017-2021 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "src/core/gpu/cl/kernels/ClCol2ImKernel.h"
+
+#include "arm_compute/core/CL/CLHelpers.h"
+#include "arm_compute/core/CL/CLKernelLibrary.h"
+#include "arm_compute/core/CL/ICLTensor.h"
+#include "arm_compute/core/CL/OpenCL.h"
+#include "arm_compute/core/Helpers.h"
+#include "arm_compute/core/utils/misc/ShapeCalculator.h"
+#include "src/core/CL/CLValidate.h"
+#include "src/core/helpers/AutoConfiguration.h"
+#include "src/core/helpers/WindowHelpers.h"
+#include "support/Cast.h"
+#include "support/StringSupport.h"
+
+#include <cmath>
+
+namespace arm_compute
+{
+using namespace misc::shape_calculator;
+namespace opencl
+{
+namespace kernels
+{
+namespace
+{
+Status validate_arguments(const ITensorInfo *src, const ITensorInfo *dst, const Size2D &convolved_dims, unsigned int num_groups)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src, dst);
+    ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(src);
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(src, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::F16, DataType::F32);
+
+    // Checks performed when output is configured
+    if(dst->total_size() != 0)
+    {
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(dst->tensor_shape(), compute_col2im_shape(*src, convolved_dims, true, num_groups));
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src, dst);
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(src, dst);
+        ARM_COMPUTE_RETURN_ERROR_ON_MSG(dst->data_layout() != DataLayout::NCHW, "Col2Im output's data layout must always be NCHW");
+    }
+
+    return Status{};
+}
+
+std::pair<Status, Window> validate_and_configure_window(ITensorInfo *src, ITensorInfo *dst, const Size2D &convolved_dims, unsigned int num_groups)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
+    // Output auto inizialitation if not yet initialized
+    auto_init_if_empty(*dst, src->clone()->set_tensor_shape(compute_col2im_shape(*src, convolved_dims, true, num_groups)).set_data_layout(DataLayout::NCHW));
+
+    constexpr unsigned int num_elems_read_per_iteration = 8;
+
+    // Configure window
+    Window win = calculate_max_window(*src, Steps(num_elems_read_per_iteration));
+
+    // Update window and padding just for the input tensor as we cannot access out-of-bounds elements in the output one
+    AccessWindowHorizontal input_access(src, 0, num_elems_read_per_iteration);
+    bool                   window_changed = update_window_and_padding(win, input_access);
+
+    Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
+    return std::make_pair(err, win);
+}
+} // namespace
+
+ClCol2ImKernel::ClCol2ImKernel()
+    : _convolved_dims()
+{
+    _type = CLKernelType::ELEMENTWISE;
+}
+
+void ClCol2ImKernel::configure(const CLCompileContext &compile_context, ITensorInfo *src, ITensorInfo *dst, const Size2D &convolved_dims, unsigned int num_groups)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
+
+    // Perform validation step
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src, dst, convolved_dims, num_groups));
+
+    _convolved_dims = convolved_dims;
+
+    const DataType data_type = src->data_type();
+
+    // Create kernel
+    CLBuildOptions build_opts;
+    build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(data_type));
+    build_opts.add_option("-DELEMENT_SIZE=" + support::cpp11::to_string(src->element_size()));
+    build_opts.add_option("-DWIDTH_INPUT=" + support::cpp11::to_string(src->dimension(0)));
+    build_opts.add_option("-DWIDTH_OUTPUT=" + support::cpp11::to_string(_convolved_dims.width));
+    build_opts.add_option("-DNUM_GROUPS=" + support::cpp11::to_string(num_groups));
+
+    _kernel = create_kernel(compile_context, "col2im", build_opts.options());
+
+    // Configure kernel window
+    auto win_config = validate_and_configure_window(src, dst, _convolved_dims, num_groups);
+    ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
+    IClKernel::configure_internal(win_config.second);
+
+    // Set config_id for enabling LWS tuning
+    _config_id = "col2im_";
+    _config_id += lower_string(string_from_data_type(src->data_type()));
+    _config_id += "_";
+    _config_id += support::cpp11::to_string(num_groups);
+    _config_id += "_";
+    _config_id += support::cpp11::to_string(src->dimension(0));
+    _config_id += "_";
+    _config_id += support::cpp11::to_string(src->dimension(1));
+    _config_id += "_";
+    _config_id += support::cpp11::to_string(dst->dimension(0));
+    _config_id += "_";
+    _config_id += support::cpp11::to_string(dst->dimension(1));
+}
+
+Status ClCol2ImKernel::validate(const ITensorInfo *src, const ITensorInfo *dst, const Size2D &convolved_dims, unsigned int num_groups)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src, dst, convolved_dims, num_groups));
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(src->clone().get(), dst->clone().get(), convolved_dims, num_groups).first);
+    return Status{};
+}
+
+void ClCol2ImKernel::run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue)
+{
+    ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
+    ARM_COMPUTE_ERROR_ON_MISMATCHING_WINDOWS(IClKernel::window(), window);
+
+    bool is_collapsed     = false;
+    bool is_collapsed_out = false;
+
+    auto src = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC));
+    auto dst = utils::cast::polymorphic_downcast<ICLTensor *>(tensors.get_tensor(TensorType::ACL_DST));
+
+    Window out_window;
+    out_window.use_tensor_dimensions(dst->info()->tensor_shape());
+
+    Window collapsed     = window.collapse_if_possible(ICLKernel::window(), Window::DimZ, &is_collapsed);
+    Window collapsed_out = out_window.collapse_if_possible(out_window, 3, &is_collapsed_out);
+
+    ARM_COMPUTE_ERROR_ON(is_collapsed != is_collapsed_out);
+
+    Window slice     = collapsed.first_slice_window_3D();
+    Window slice_out = collapsed_out.first_slice_window_4D();
+    do
+    {
+        // Set inputs
+        unsigned int idx = 0;
+        add_3D_tensor_argument(idx, src, slice);
+        add_4D_tensor_argument(idx, dst, slice_out);
+        enqueue(queue, *this, slice, lws_hint());
+    }
+    while(collapsed.slide_window_slice_3D(slice) && collapsed_out.slide_window_slice_4D(slice_out));
+}
+} // namespace kernels
+} // namespace opencl
+} // namespace arm_compute
diff --git a/src/core/gpu/cl/kernels/ClCol2ImKernel.h b/src/core/gpu/cl/kernels/ClCol2ImKernel.h
new file mode 100644
index 0000000..42d0a96
--- /dev/null
+++ b/src/core/gpu/cl/kernels/ClCol2ImKernel.h
@@ -0,0 +1,89 @@
+/*
+ * Copyright (c) 2017-2021 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef ARM_COMPUTE_CL_COL2IM_KERNEL_H
+#define ARM_COMPUTE_CL_COL2IM_KERNEL_H
+
+#include "arm_compute/core/Size2D.h"
+#include "src/core/common/Macros.h"
+#include "src/core/gpu/cl/ClCompileContext.h"
+#include "src/core/gpu/cl/IClKernel.h"
+
+namespace arm_compute
+{
+namespace opencl
+{
+namespace kernels
+{
+/** Interface for the col2im reshaping kernel.
+ *
+ * Rearranges each matrix column into image blocks. It's the inverse operation of @ref CLIm2ColKernel.
+ *
+ * For example, a vector of 9 elements can be reshaped to a block(image) of 3x3:
+ *
+ * @f[
+ * \left( \begin{array}{ccccccccc}
+ * a0 & a1 & a2 & a3 & a4 & a5 & a6 & a7 & a8 \\
+ * \end{array} \right)
+ * \rightarrow
+ * \left( \begin{array}{ccc}
+ * a0 & a1 & a2 \\
+ * a3 & a4 & a5 \\
+ * a6 & a7 & a8 \\
+ * \end{array} \right)
+ * @f]
+ */
+class ClCol2ImKernel : public IClKernel
+{
+public:
+    /** Default constructor */
+    ClCol2ImKernel();
+    ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClCol2ImKernel);
+    /** Set the input and output of the kernel.
+     *
+     * @param[in]  compile_context The compile context to be used.
+     * @param[in]  src             The input tensor info to convert. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32
+     * @param[out] dst             The output tensor info. 3 lower dimensions represent a single output [width, height, OFM],
+     *                             while the rest represent batch of outputs. Data types supported: Same as @p input. Data layout: NCHW
+     * @param[in]  convolved_dims  Output convolved dimensions.
+     * @param[in]  num_groups      (Optional) Number of groups when performing a grouped convolution
+     */
+    void configure(const CLCompileContext &compile_context, ITensorInfo *src, ITensorInfo *dst, const Size2D &convolved_dims, unsigned int num_groups = 1);
+    /** Static function to check if given info will lead to a valid configuration
+     *
+     * Similar to ClCol2ImKernel::configure()
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *src, const ITensorInfo *dst, const Size2D &convolved_dims, unsigned int num_groups = 1);
+
+    // Inherited methods overridden:
+    void run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue) override;
+
+public:
+    Size2D _convolved_dims;
+};
+} // namespace kernels
+} // namespace opencl
+} // namespace arm_compute
+#endif /*ARM_COMPUTE_CL_COL2IM_KERNEL_H */
diff --git a/src/runtime/CL/functions/CLGEMMConvolutionLayer.cpp b/src/runtime/CL/functions/CLGEMMConvolutionLayer.cpp
index cef8ad5..f0f45a8 100644
--- a/src/runtime/CL/functions/CLGEMMConvolutionLayer.cpp
+++ b/src/runtime/CL/functions/CLGEMMConvolutionLayer.cpp
@@ -30,9 +30,9 @@
 #include "arm_compute/core/utils/misc/ShapeCalculator.h"
 #include "arm_compute/core/utils/quantization/AsymmHelpers.h"
 #include "arm_compute/runtime/CL/CLScheduler.h"
-#include "src/core/CL/kernels/CLCol2ImKernel.h"
 #include "src/core/CL/kernels/CLIm2ColKernel.h"
 #include "src/core/CL/kernels/CLWeightsReshapeKernel.h"
+#include "src/core/gpu/cl/kernels/ClCol2ImKernel.h"
 #include "src/core/helpers/AutoConfiguration.h"
 #include "support/Cast.h"
 
@@ -106,8 +106,8 @@
 
 CLGEMMConvolutionLayer::CLGEMMConvolutionLayer(std::shared_ptr<IMemoryManager> memory_manager, IWeightsManager *weights_manager)
     : _memory_group(memory_manager), _weights_manager(weights_manager), _reshape_weights(), _reshape_weights_managed(), _im2col_kernel(std::make_unique<CLIm2ColKernel>()), _mm_gemm(memory_manager,
-            weights_manager), _mm_gemmlowp(memory_manager), _col2im_kernel(std::make_unique<CLCol2ImKernel>()), _activationlayer_function(), _original_weights(nullptr), _im2col_output(), _weights_reshaped(),
-      _gemm_output(), _skip_im2col(false), _skip_col2im(false), _is_quantized(false), _fuse_activation(true), _is_prepared(false)
+            weights_manager), _mm_gemmlowp(memory_manager), _col2im_kernel(nullptr), _activationlayer_function(), _original_weights(nullptr), _gemm_output_to_use(nullptr), _output(nullptr), _im2col_output(),
+      _weights_reshaped(), _gemm_output(), _skip_im2col(false), _skip_col2im(false), _is_quantized(false), _fuse_activation(true), _is_prepared(false)
 {
 }
 
@@ -238,7 +238,6 @@
 
     // Set the GPU target for im2col and col2im
     _im2col_kernel->set_target(CLScheduler::get().target());
-    _col2im_kernel->set_target(CLScheduler::get().target());
 
     const ICLTensor *gemm_input_to_use  = input;
     ICLTensor       *gemm_output_to_use = output;
@@ -395,9 +394,14 @@
 
     if(!_skip_col2im)
     {
+        // Set the GPU target for col2im
+        _col2im_kernel = std::make_unique<opencl::kernels::ClCol2ImKernel>();
+        _col2im_kernel->set_target(CLScheduler::get().target());
         // Configure and tune Col2Im
-        _col2im_kernel->configure(compile_context, gemm_output_to_use, output, Size2D(conv_w, conv_h), num_groups);
+        _col2im_kernel->configure(compile_context, gemm_output_to_use->info(), output->info(), Size2D(conv_w, conv_h), num_groups);
         CLScheduler::get().tune_kernel_static(*_col2im_kernel.get());
+        _gemm_output_to_use = gemm_output_to_use;
+        _output             = output;
     }
 
     if(!_skip_col2im)
@@ -595,7 +599,7 @@
     // Validate Col2Im
     if(!skip_col2im)
     {
-        ARM_COMPUTE_RETURN_ON_ERROR(CLCol2ImKernel::validate(gemm_output_to_use, output, Size2D(conv_w, conv_h), num_groups));
+        ARM_COMPUTE_RETURN_ON_ERROR(opencl::kernels::ClCol2ImKernel::validate(gemm_output_to_use, output, Size2D(conv_w, conv_h), num_groups));
     }
 
     //Validate Activation Layer
@@ -634,7 +638,12 @@
     // Reshape output matrix
     if(!_skip_col2im)
     {
-        CLScheduler::get().enqueue(*_col2im_kernel.get(), false);
+        ITensorPack pack =
+        {
+            { TensorType::ACL_SRC, _gemm_output_to_use },
+            { TensorType::ACL_DST, _output }
+        };
+        CLScheduler::get().enqueue_op(*_col2im_kernel.get(), pack, false);
     }
 
     //Run Activation Layer if we cannot fuse in GEMM