diff --git a/src/core/CL/ICLKernel.cpp b/src/core/CL/ICLKernel.cpp
index 9bbc710..109a076 100644
--- a/src/core/CL/ICLKernel.cpp
+++ b/src/core/CL/ICLKernel.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-2021 Arm Limited.
+ * Copyright (c) 2016-2022 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -173,6 +173,7 @@
 template void ICLKernel::add_tensor_argument<2>(unsigned &idx, const ICLTensor *tensor, const Window &window);
 template void ICLKernel::add_tensor_argument<3>(unsigned &idx, const ICLTensor *tensor, const Window &window);
 template void ICLKernel::add_tensor_argument<4>(unsigned &idx, const ICLTensor *tensor, const Window &window);
+template void ICLKernel::add_tensor_argument<5>(unsigned &idx, const ICLTensor *tensor, const Window &window);
 #endif /* DOXYGEN_SKIP_THIS */
 
 void ICLKernel::set_target(cl::Device &device)
diff --git a/src/core/CL/ICLKernel.h b/src/core/CL/ICLKernel.h
index 4c8028e..046679e 100644
--- a/src/core/CL/ICLKernel.h
+++ b/src/core/CL/ICLKernel.h
@@ -238,6 +238,16 @@
     {
         add_tensor_argument<4>(idx, tensor, window);
     }
+    /** Add the passed 5D tensor's parameters to the object's kernel's arguments starting from the index idx.
+     *
+     * @param[in,out] idx    Index at which to start adding the tensor's arguments. Will be incremented by the number of kernel arguments set.
+     * @param[in]     tensor Tensor to set as an argument of the object's kernel.
+     * @param[in]     window Window the kernel will be executed on.
+     */
+    void add_5D_tensor_argument(unsigned int &idx, const ICLTensor *tensor, const Window &window)
+    {
+        add_tensor_argument<5>(idx, tensor, window);
+    }
 
     /** Add the passed NHW 3D tensor's parameters to the object's kernel's arguments by passing strides, dimensions and the offset to the first valid element in bytes.
      *
diff --git a/src/core/CL/cl_kernels/helpers.h b/src/core/CL/cl_kernels/helpers.h
index bfb693e..4018c40 100644
--- a/src/core/CL/cl_kernels/helpers.h
+++ b/src/core/CL/cl_kernels/helpers.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-2021 Arm Limited.
+ * Copyright (c) 2016-2022 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -831,6 +831,20 @@
     uint        name##_step_w,   \
     uint        name##_offset_first_element_in_bytes
 
+#define TENSOR5D_DECLARATION(name)   \
+    __global uchar *name##_ptr,      \
+    uint        name##_stride_x, \
+    uint        name##_step_x,   \
+    uint        name##_stride_y, \
+    uint        name##_step_y,   \
+    uint        name##_stride_z, \
+    uint        name##_step_z,   \
+    uint        name##_stride_w, \
+    uint        name##_step_w,   \
+    uint        name##_stride_v, \
+    uint        name##_step_v,   \
+    uint        name##_offset_first_element_in_bytes
+
 #define CONVERT_TO_VECTOR_STRUCT(name) \
     update_vector_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x)
 
diff --git a/src/core/CL/cl_kernels/nhwc/pooling_3d_layer.cl b/src/core/CL/cl_kernels/nhwc/pooling_3d_layer.cl
new file mode 100644
index 0000000..7c64143
--- /dev/null
+++ b/src/core/CL/cl_kernels/nhwc/pooling_3d_layer.cl
@@ -0,0 +1,190 @@
+/*
+ * Copyright (c) 2022 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 "helpers.h"
+#include "tile_helpers.h" // Needed for GET_SPATIAL_IDX()
+
+#if defined(POOL_AVG) || defined(POOL_L2)
+#define POOL_OP(x, y) ((x) + (y))
+#else /* defined(POOL_AVG) || defined(POOL_L2) */
+#define POOL_OP(x, y) (fmax((x), (y)))
+#endif /* defined(POOL_AVG) || defined(POOL_L2) */
+
+#define SQRT_OP(x) sqrt((x))
+
+#if defined(VEC_SIZE) && defined(VEC_SIZE_LEFTOVER) && defined(SRC_WIDTH) && defined(SRC_HEIGHT) && defined(SRC_DEPTH) && defined(DST_CHANNELS) && defined(DST_HEIGHT) && defined(DST_DEPTH) && defined(DST_BATCH_SIZE) && defined(ACC_DATA_TYPE)
+
+#if defined(POOL_SIZE_X) && defined(POOL_SIZE_Y) && defined(POOL_SIZE_Z)
+
+/** Performs 3d pooling layer of size equal to MxNXD. This OpenCL kernel can perform the following pooling types:
+ * -# max, -DPOOL_MAX must be passed at compile time
+ * -# average, -DPOOL_AVG must be passed at compile time. If padding has to be excluded, -DEXCLUDE_PADDING should be passed at compile time
+ * -# l2 normalisation, -DPOOL_L2 must be passed at compile time
+ *
+ * @note Datatype must be passed at compile type using -DDATA_TYPE e.g. -DDATA_TYPE=half. Supported data types are F32/F16
+ * @note Accumulation data type must be passed at compile time using -DACC_DATA_TYPE e.g. -DACC_DATA_TYPE=float
+ * @note If -DFP_MIXED_PRECISION is passed at compile time, the kernel will use F32 for the partial result
+ * @note Pool size must be passed at compile time using -DPOOL_SIZE_X, -DPOOL_SIZE_Y, and -DPOOL_SIZE_Z. e.g. -DPOOL_SIZE_X=4, -DPOOL_SIZE_Y=4, -DPOOL_SIZE_Z=2
+ * @note Input tensor width, height and depth must be passed at compile time using -DSRC_WIDTH, -DSRC_HEIGHT, and -DSRC_DEPTH
+ * @note Output tensor height, channels, depth, and batch size must be passed at compile time using -DDST_HEIGHT, -DDST_CHANNELS, -DDST_DEPTH, and -DDST_BATCH_SIZE
+ * @note Pool strides must be passed at compile time using -DSTRIDE_X, -DSTRIDE_Y and -DSTRIDE_Z which are the steps of the window along the x, y and z directions
+ * @note Pool pads must be passed at compile time using -DPAD_X, -DPAD_Y, -DPAD_Z
+ * @note Vector size must be passed at compile time using -DVEC_SIZE=size. e.g. -DVEC_SIZE=16
+ * @note Leftover vector size must be passed at compile time using -DVEC_SIZE_LEFTOVER. e.g. -DVEC_SIZE_LEFTOVER=3. It is defined as the remainder between the input's first dimension and VEC_SIZE
+ * @note The initial value for the pooling operation must be passed at compile time using -DINITIAL_VALUE e.g. -DINITIAL_VALUE=0
+ *
+ * @param[in]  input_ptr                            Pointer to the source tensor. Supported data types: F32/F16
+ * @param[in]  input_stride_x                       Stride of the source tensor in X dimension (in bytes)
+ * @param[in]  input_step_x                         input_stride_x * number of elements along X processed per workitem(in bytes)
+ * @param[in]  input_stride_y                       Stride of the source tensor in Y dimension (in bytes)
+ * @param[in]  input_step_y                         input_stride_y * number of elements along Y processed per workitem(in bytes)
+ * @param[in]  input_stride_z                       Stride of the source tensor in Z dimension (in bytes)
+ * @param[in]  input_step_z                         input_stride_z * number of elements along Z processed per workitem(in bytes)
+ * @param[in]  input_stride_w                       Stride of the source tensor in W dimension (in bytes)
+ * @param[in]  input_step_w                         input_stride_w * number of elements along W processed per workitem(in bytes)
+ * @param[in]  input_stride_v                       Stride of the source tensor in V dimension (in bytes)
+ * @param[in]  input_step_v                         input_stride_v * number of elements along V processed per workitem(in bytes)
+ * @param[in]  input_offset_first_element_in_bytes  The offset of the first element in the source tensor
+ * @param[out] output_ptr                           Pointer to the destination tensor. Supported data types: same as @p input_ptr
+ * @param[in]  output_stride_x                      Stride of the destination tensor in X dimension (in bytes)
+ * @param[in]  output_step_x                        output_stride_x * number of elements along X processed per workitem(in bytes)
+ * @param[in]  output_stride_y                      Stride of the destination tensor in Y dimension (in bytes)
+ * @param[in]  output_step_y                        output_stride_y * number of elements along Y processed per workitem(in bytes)
+ * @param[in]  output_stride_z                      Stride of the destination tensor in Z dimension (in bytes)
+ * @param[in]  output_step_z                        output_stride_z * number of elements along Z processed per workitem(in bytes)
+ * @param[in]  output_stride_w                      Stride of the destination tensor in W dimension (in bytes)
+ * @param[in]  output_step_w                        output_stride_w * number of elements along W processed per workitem(in bytes)
+ * @param[in]  output_stride_v                      Stride of the destination tensor in V dimension (in bytes)
+ * @param[in]  output_step_v                        output_stride_v * number of elements along V processed per workitem(in bytes)
+ * @param[in]  output_offset_first_element_in_bytes The offset of the first element in the destination tensor
+ */
+__kernel void pooling_3d_layer_MxN_ndhwc(
+    TENSOR5D_DECLARATION(input),
+    TENSOR5D_DECLARATION(output))
+{
+    // Note: If C is not multiple of VEC_SIZE, we shift back of VEC_SIZE_LEFTOVER elements to compute the leftover elements for get_global_id(0) == 0
+    // Note: If C is less than VEC_SIZE, VEC_SIZE should be SHRINKED to the closest smaller VEC_SIZE. This operation is performed on the host side
+    int idx_out_c = GET_SPATIAL_IDX(0, VEC_SIZE, VEC_SIZE_LEFTOVER);
+    int idx_out_w = GET_SPATIAL_IDX(1, 1, 0);
+
+    // The depth size dimension and the batch size dimension are collapsed over the height dimension
+    int idx_out_h = GET_SPATIAL_IDX(2, 1, 0) % DST_HEIGHT;
+    int idx_out_d = (GET_SPATIAL_IDX(2, 1, 0) / DST_HEIGHT) % DST_DEPTH;
+    int idx_out_n = (GET_SPATIAL_IDX(2, 1, 0) / DST_HEIGHT) / DST_DEPTH;
+
+    __global unsigned char *in_base_ptr = input_ptr + input_offset_first_element_in_bytes + idx_out_c * sizeof(DATA_TYPE) + idx_out_n * input_stride_v;
+
+    __global unsigned char *out_base_ptr = output_ptr + output_offset_first_element_in_bytes + idx_out_c * sizeof(DATA_TYPE) + idx_out_w * output_stride_y + idx_out_h * output_stride_z + idx_out_d *
+                                           output_stride_w + idx_out_n * output_stride_v;
+
+    VEC_DATA_TYPE(ACC_DATA_TYPE, VEC_SIZE)
+    res0 = INITIAL_VALUE;
+
+    int idx_in_w = idx_out_w * STRIDE_X - (int)PAD_X;
+    int idx_in_h = idx_out_h * STRIDE_Y - (int)PAD_Y;
+    int idx_in_d = idx_out_d * STRIDE_Z - (int)PAD_Z;
+
+    // The start of width to consider in calculation should exclude padding
+    int pool_x_s = max((int)0, -idx_in_w);
+    // Assumed Symmetric Padding (left padding = right padding = PAD_X), the filter end should be either the pool width or what is remaining from current pos to the (src width + pad right)
+    int pool_x_e = min((int)POOL_SIZE_X, (int)SRC_WIDTH + PAD_X - idx_in_w);
+    int pool_y_s = max((int)0, -idx_in_h);
+    int pool_y_e = min((int)POOL_SIZE_Y, (int)SRC_HEIGHT + PAD_Y - idx_in_h);
+    int pool_z_s = max((int)0, -idx_in_d);
+    int pool_z_e = min((int)POOL_SIZE_Z, (int)SRC_DEPTH + PAD_Z - idx_in_d);
+
+    // The filter size with all padding in all directions considered.
+    int filter_size = pool_z_e * pool_y_e * pool_x_e;
+
+    // The end of width to consider in calculation should exclude PAD_X
+    pool_x_e = min(pool_x_e, SRC_WIDTH - idx_in_w);
+    pool_y_e = min(pool_y_e, SRC_HEIGHT - idx_in_h);
+    pool_z_e = min(pool_z_e, SRC_DEPTH - idx_in_d);
+
+#if defined(EXCLUDE_PADDING)
+    filter_size = (pool_z_e - pool_z_s) * (pool_y_e - pool_y_s) * (pool_x_e - pool_x_s);
+#endif  // defined(EXCLUDE_PADDING)
+
+#if POOL_SIZE_X == SRC_WIDTH && POOL_SIZE_Y == SRC_HEIGHT && POOL_SIZE_Z == SRC_DEPTH && PAD_X == 0 && PAD_Y == 0 && PAD_Z == 0
+    // Global pooling path
+    for(int z = 0; z < POOL_SIZE_Z; ++z)
+    {
+        int depth_offset_src = (z + idx_in_d) * input_stride_w;
+        for(int y = 0; y < POOL_SIZE_Y; ++y)
+        {
+            int height_offset_src = (y + idx_in_h) * input_stride_z;
+#pragma unroll 8
+            for(int x = 0; x < POOL_SIZE_X; ++x)
+            {
+                int width_offset_src = (x + idx_in_w) * input_stride_y;
+#else // POOL_SIZE_X == SRC_WIDTH && POOL_SIZE_Y == SRC_HEIGHT && POOL_SIZE_Z == SRC_DEPTH && PAD_X == 0 && PAD_Y == 0 && PAD_Z == 0
+    for(int z = pool_z_s; z < pool_z_e; ++z)
+    {
+        int depth_offset_src = (z + idx_in_d) * input_stride_w;
+        for(int y = pool_y_s; y < pool_y_e; ++y)
+        {
+            int height_offset_src = (y + idx_in_h) * input_stride_z;
+#pragma unroll 8
+            for(int x = pool_x_s; x < pool_x_e; ++x)
+            {
+                int width_offset_src = (x + idx_in_w) * input_stride_y;
+#endif // POOL_SIZE_X == SRC_WIDTH && POOL_SIZE_Y == SRC_HEIGHT && POOL_SIZE_Z == SRC_DEPTH && PAD_X == 0 && PAD_Y == 0 && PAD_Z == 0
+                VEC_DATA_TYPE(ACC_DATA_TYPE, VEC_SIZE)
+                data0;
+#if defined(FP_MIXED_PRECISION)
+                // In case of FP_MIXED_PRECISION, ACC_DATA_TYPE is != DATA_TYPE
+                data0 = CONVERT(VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(in_base_ptr + width_offset_src + height_offset_src + depth_offset_src)),
+                                VEC_DATA_TYPE(ACC_DATA_TYPE, VEC_SIZE));
+#else  // defined(FP_MIXED_PRECISION)
+                data0 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(in_base_ptr + width_offset_src + height_offset_src + depth_offset_src));
+#endif // defined(FP_MIXED_PRECISION)
+
+#if defined(POOL_L2)
+                // Raise to power of 2 for L2 Pooling
+                data0 *= data0;
+#endif // defined(POOL_L2)
+                res0 = POOL_OP(res0, data0);
+            }
+        }
+    }
+
+#if defined(POOL_AVG) || defined(POOL_L2)
+    res0 /= (VEC_DATA_TYPE(ACC_DATA_TYPE, VEC_SIZE))filter_size;
+#endif // defined(POOL_AVG) || defined(POOL_L2)
+
+#if defined(POOL_L2)
+    // Take square root of the result in L2 pooling
+    res0 = SQRT_OP(res0);
+#endif // defined(POOL_L2)
+
+    // Store result
+#if defined(FP_MIXED_PRECISION)
+    VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE)
+    res_converted0 = CONVERT(res0, VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE));
+    STORE_VECTOR_SELECT(res_converted, DATA_TYPE, out_base_ptr, VEC_SIZE, VEC_SIZE_LEFTOVER, (VEC_SIZE_LEFTOVER != 0) && get_global_id(0) == 0);
+#else  // defined(FP_MIXED_PRECISION)
+    STORE_VECTOR_SELECT(res, DATA_TYPE, out_base_ptr, VEC_SIZE, VEC_SIZE_LEFTOVER, (VEC_SIZE_LEFTOVER != 0) && get_global_id(0) == 0);
+#endif // defined(FP_MIXED_PRECISION)
+}
+#endif // defined(POOL_SIZE_X) && defined(POOL_SIZE_Y) && defined(POOL_SIZE_Z)
+#endif // defined(VEC_SIZE) && defined(VEC_SIZE_LEFTOVER) && defined(SRC_WIDTH) && defined(SRC_HEIGHT) && defined(SRC_DEPTH) && defined(DST_CHANNELS) && defined(DST_HEIGHT) && defined(DST_DEPTH) && defined(DST_BATCH_SIZE) && defined(ACC_DATA_TYPE)
diff --git a/src/core/Utils.cpp b/src/core/Utils.cpp
index ef16cb5..904362e 100644
--- a/src/core/Utils.cpp
+++ b/src/core/Utils.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-2021 Arm Limited.
+ * Copyright (c) 2016-2022 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -244,6 +244,19 @@
     return pool_le_padding_x || pool_le_padding_y;
 }
 
+bool is_pool_3d_region_entirely_outside_input(const Pooling3dLayerInfo &info)
+{
+    if(info.is_global_pooling || info.pool_size.x() == 0 || info.pool_size.y() == 0 || info.pool_size.z() == 0)
+    {
+        return false;
+    }
+    const auto ps                = info.padding;
+    const auto pool_le_padding_x = info.pool_size.x() <= std::max({ ps.left, ps.right });
+    const auto pool_le_padding_y = info.pool_size.y() <= std::max({ ps.top, ps.bottom });
+    const auto pool_le_padding_z = info.pool_size.z() <= std::max({ ps.front, ps.back });
+    return pool_le_padding_x || pool_le_padding_y || pool_le_padding_z;
+}
+
 const std::string &string_from_gemmlowp_output_stage(GEMMLowpOutputStageType output_stage)
 {
     static std::map<GEMMLowpOutputStageType, const std::string> output_stage_map =
@@ -474,6 +487,42 @@
     return std::make_pair(static_cast<int>(w), static_cast<int>(h));
 }
 
+std::tuple<int, int, int> scaled_3d_dimensions_signed(int width, int height, int depth,
+                                                      int kernel_width, int kernel_height, int kernel_depth,
+                                                      const Pooling3dLayerInfo &pool3d_info)
+{
+    const int pad_left   = pool3d_info.padding.left;
+    const int pad_top    = pool3d_info.padding.top;
+    const int pad_right  = pool3d_info.padding.right;
+    const int pad_bottom = pool3d_info.padding.bottom;
+    const int pad_front  = pool3d_info.padding.front;
+    const int pad_back   = pool3d_info.padding.back;
+    const int stride_x   = pool3d_info.stride.x();
+    const int stride_y   = pool3d_info.stride.y();
+    const int stride_z   = pool3d_info.stride.z();
+    int       w          = 0;
+    int       h          = 0;
+    int       d          = 0;
+
+    switch(pool3d_info.round_type)
+    {
+        case DimensionRoundingType::FLOOR:
+            w = static_cast<int>(std::floor((static_cast<float>(width + pad_left + pad_right - kernel_width) / stride_x) + 1));
+            h = static_cast<int>(std::floor((static_cast<float>(height + pad_top + pad_bottom - kernel_height) / stride_y) + 1));
+            d = static_cast<int>(std::floor((static_cast<float>(depth + pad_front + pad_back - kernel_depth) / stride_z) + 1));
+            break;
+        case DimensionRoundingType::CEIL:
+            w = static_cast<int>(std::ceil((static_cast<float>(width + pad_left + pad_right - kernel_width) / stride_x) + 1));
+            h = static_cast<int>(std::ceil((static_cast<float>(height + pad_top + pad_bottom - kernel_height) / stride_y) + 1));
+            d = static_cast<int>(std::ceil((static_cast<float>(depth + pad_front + pad_back - kernel_depth) / stride_z) + 1));
+            break;
+        default:
+            ARM_COMPUTE_ERROR("Unsupported rounding type");
+    }
+
+    return std::make_tuple(static_cast<int>(w), static_cast<int>(h), static_cast<int>(d));
+}
+
 bool needs_serialized_reduction(ReductionOperation op, DataType dt, unsigned int axis)
 {
     const bool is_min_max        = (op == ReductionOperation::MAX || op == ReductionOperation::MIN);
diff --git a/src/gpu/cl/ClKernelLibrary.cpp b/src/gpu/cl/ClKernelLibrary.cpp
index bab5342..a5d37f4 100644
--- a/src/gpu/cl/ClKernelLibrary.cpp
+++ b/src/gpu/cl/ClKernelLibrary.cpp
@@ -437,6 +437,7 @@
     { "pooling_layer_MxN_nhwc", "nhwc/pooling_layer.cl" },
     { "pooling_layer_2x2_nhwc", "nhwc/pooling_layer.cl" },
     { "pooling_layer_MxN_quantized_nhwc", "nhwc/pooling_layer_quantized.cl" },
+    { "pooling_3d_layer_MxN_ndhwc", "nhwc/pooling_3d_layer.cl" },
     { "reorg_layer_nhwc", "nhwc/reorg_layer.cl" },
     { "scale_nearest_neighbour_nhwc", "nhwc/scale.cl" },
     { "scale_bilinear_nhwc", "nhwc/scale.cl" },
@@ -882,6 +883,10 @@
 #include "./cl_kernels/nhwc/pooling_layer.clembed"
     },
     {
+        "nhwc/pooling_3d_layer.cl",
+#include "./cl_kernels/nhwc/pooling_3d_layer.clembed"
+    },
+    {
         "nhwc/pooling_layer_quantized.cl",
 #include "./cl_kernels/nhwc/pooling_layer_quantized.clembed"
     },
diff --git a/src/gpu/cl/kernels/ClPool3dKernel.cpp b/src/gpu/cl/kernels/ClPool3dKernel.cpp
new file mode 100644
index 0000000..929ccf7
--- /dev/null
+++ b/src/gpu/cl/kernels/ClPool3dKernel.cpp
@@ -0,0 +1,225 @@
+/*
+ * Copyright (c) 2022 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/gpu/cl/kernels/ClPool3dKernel.h"
+
+#include "arm_compute/core/CL/ICLTensor.h"
+#include "arm_compute/core/TensorInfo.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 "utils/TypePrinter.h"
+
+namespace arm_compute
+{
+namespace opencl
+{
+namespace kernels
+{
+using namespace arm_compute::misc::shape_calculator;
+
+namespace
+{
+Status validate_arguments(const ITensorInfo *src, const ITensorInfo *dst, const Pooling3dLayerInfo &pool_info)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src, dst);
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG(src->data_layout() != DataLayout::NDHWC, "Only NDHWC layout supported");
+
+    ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(src);
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG((pool_info.stride.x() == 0 || pool_info.stride.y() == 0 || pool_info.stride.z() == 0), "Strides cannot be zero.");
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(src, 1, DataType::F16, DataType::F32);
+
+    const auto   data_layout       = src->data_layout();
+    const int    idx_width         = get_data_layout_dimension_index(data_layout, DataLayoutDimension::WIDTH);
+    const int    idx_height        = get_data_layout_dimension_index(data_layout, DataLayoutDimension::HEIGHT);
+    const int    idx_depth         = get_data_layout_dimension_index(data_layout, DataLayoutDimension::DEPTH);
+    const bool   is_global_pooling = pool_info.is_global_pooling;
+    const unsigned int pool_size_x       = is_global_pooling ? src->dimension(idx_width) : pool_info.pool_size.width;
+    const unsigned int pool_size_y       = is_global_pooling ? src->dimension(idx_height) : pool_info.pool_size.height;
+    const unsigned int pool_size_z       = is_global_pooling ? src->dimension(idx_depth) : pool_info.pool_size.depth;
+    int          output_width      = 0;
+    int          output_height     = 0;
+    int          output_depth      = 0;
+
+    bool round_type_ceil_with_asymm_padding = (pool_info.round_type == DimensionRoundingType::CEIL) && (!is_symmetric(pool_info.padding));
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG(round_type_ceil_with_asymm_padding, "Cannot use dimension round type CEIL when padding is asymmetric.");
+
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG(is_pool_3d_region_entirely_outside_input(pool_info), "Pooling region that is entirely outside input tensor is unsupported");
+    std::tie(output_width, output_height, output_depth) = scaled_3d_dimensions_signed(src->tensor_shape()[idx_width], src->tensor_shape()[idx_height],
+                                                                                      src->tensor_shape()[idx_depth], pool_size_x, pool_size_y,
+                                                                                      pool_size_z, pool_info);
+
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG((output_width < 1 || output_height < 1 || output_depth < 1), "Calculated output dimension size is invalid");
+    // Checks performed when dst is configured
+    if(dst->total_size() != 0)
+    {
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src, dst);
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(src, dst);
+        TensorInfo out_info(TensorInfo(compute_pool3d_shape(src->tensor_shape(), pool_info), 1, dst->data_type()));
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(dst, &out_info);
+    }
+
+    return Status{};
+}
+} // namespace
+
+ClPool3dKernel::ClPool3dKernel()
+{
+    _type = CLKernelType::POOL;
+}
+
+void ClPool3dKernel::configure(const ClCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst, const Pooling3dLayerInfo &pool_info)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src, dst, pool_info));
+    auto padding_info = get_padding_info({ src, dst });
+
+    // Auto init if empty
+    TensorShape out_shape = compute_pool3d_shape(src->tensor_shape(), pool_info);
+    auto_init_if_empty(*dst, src->clone()->set_tensor_shape(out_shape));
+
+    // Set instance variables
+    _pool_info   = pool_info;
+    _data_layout = src->data_layout();
+
+    _num_elems_processed_per_iteration = (dst->data_type() == DataType::F32) ? 2 : 4;
+    _num_elems_processed_per_iteration = adjust_vec_size(_num_elems_processed_per_iteration, dst->dimension(0));
+
+    const PoolingType pool_type       = pool_info.pool_type;
+    const int         idx_width       = get_data_layout_dimension_index(_data_layout, DataLayoutDimension::WIDTH);
+    const int         idx_height      = get_data_layout_dimension_index(_data_layout, DataLayoutDimension::HEIGHT);
+    const int         idx_depth       = get_data_layout_dimension_index(_data_layout, DataLayoutDimension::DEPTH);
+    const int         idx_channel     = get_data_layout_dimension_index(_data_layout, DataLayoutDimension::CHANNEL);
+    const int         idx_batch_size  = get_data_layout_dimension_index(_data_layout, DataLayoutDimension::BATCHES);
+    const int         pool_size_x     = pool_info.is_global_pooling ? src->dimension(idx_width) : pool_info.pool_size.width;
+    const int         pool_size_y     = pool_info.is_global_pooling ? src->dimension(idx_height) : pool_info.pool_size.height;
+    const int         pool_size_z     = pool_info.is_global_pooling ? src->dimension(idx_depth) : pool_info.pool_size.depth;
+    const bool        exclude_padding = pool_info.exclude_padding;
+    const int         pool_stride_x   = pool_info.stride.x();
+    const int         pool_stride_y   = pool_info.stride.y();
+    const int         pool_stride_z   = pool_info.stride.z();
+    const int         pool_pad_top    = pool_info.padding.top;
+    const int         pool_pad_left   = pool_info.padding.left;
+    const int         pool_pad_front  = pool_info.padding.front;
+    const DataType    data_type       = src->data_type();
+
+    // Set build options
+    CLBuildOptions build_opts;
+    build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(_num_elems_processed_per_iteration));
+    build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(data_type));
+    build_opts.add_option("-DPOOL_" + string_from_pooling_type(pool_type));
+    build_opts.add_option("-DSTRIDE_X=" + support::cpp11::to_string(pool_stride_x));
+    build_opts.add_option("-DSTRIDE_Y=" + support::cpp11::to_string(pool_stride_y));
+    build_opts.add_option("-DSTRIDE_Z=" + support::cpp11::to_string(pool_stride_z));
+    build_opts.add_option("-DPAD_X=" + support::cpp11::to_string(pool_pad_left));
+    build_opts.add_option("-DPAD_Y=" + support::cpp11::to_string(pool_pad_top));
+    build_opts.add_option("-DPAD_Z=" + support::cpp11::to_string(pool_pad_front));
+    build_opts.add_option("-DPOOL_SIZE_X=" + support::cpp11::to_string(pool_size_x));
+    build_opts.add_option("-DPOOL_SIZE_Y=" + support::cpp11::to_string(pool_size_y));
+    build_opts.add_option("-DPOOL_SIZE_Z=" + support::cpp11::to_string(pool_size_z));
+    build_opts.add_option("-DSRC_WIDTH=" + support::cpp11::to_string(src->dimension(idx_width)));
+    build_opts.add_option("-DSRC_HEIGHT=" + support::cpp11::to_string(src->dimension(idx_height)));
+    build_opts.add_option("-DSRC_DEPTH=" + support::cpp11::to_string(src->dimension(idx_depth)));
+
+    // Set the initial value for the pooling operation accordingly with the data type
+    if(pool_type == PoolingType::MAX)
+    {
+        build_opts.add_option("-DINITIAL_VALUE=" + float_to_string_with_full_precision(std::numeric_limits<float>::lowest()));
+    }
+    else
+    {
+        // Pool AVG and Pool L2 initial value
+        build_opts.add_option("-DINITIAL_VALUE=0");
+    }
+    // Create kernel
+    // Floating point mixed precision is support on F16 only
+    const auto use_fp_mixed_precision = (data_type == DataType::F16) && pool_info.fp_mixed_precision && pool_type != PoolingType::MAX;
+
+    // Wider accumulation is required to avoid accuracy loss
+    // Case 1: Floating point mixed precision (fp16 src data and fp32 accumulation)
+    DataType acc_data_type = data_type;
+    if(use_fp_mixed_precision)
+    {
+        acc_data_type = DataType::F32;
+    }
+    build_opts.add_option("-DACC_DATA_TYPE=" + get_cl_type_from_data_type(acc_data_type));
+    build_opts.add_option_if(use_fp_mixed_precision, "-DFP_MIXED_PRECISION");
+    build_opts.add_option_if(exclude_padding, "-DEXCLUDE_PADDING");
+    build_opts.add_option("-DDST_HEIGHT=" + support::cpp11::to_string(dst->dimension(idx_height)));
+    build_opts.add_option("-DDST_DEPTH=" + support::cpp11::to_string(dst->dimension(idx_depth)));
+    build_opts.add_option("-DDST_CHANNELS=" + support::cpp11::to_string(dst->dimension(idx_channel)));
+    build_opts.add_option("-DDST_BATCH_SIZE=" + support::cpp11::to_string(dst->dimension(idx_batch_size)));
+    build_opts.add_option("-DVEC_SIZE_LEFTOVER=" + support::cpp11::to_string(src->dimension(0) % _num_elems_processed_per_iteration));
+    std::string kernel_name = "pooling_3d_layer_MxN_ndhwc";
+    _kernel                 = create_kernel(compile_context, kernel_name, build_opts.options());
+
+    // Configure kernel window
+    Window win = calculate_max_window(*dst, Steps(_num_elems_processed_per_iteration));
+    ICLKernel::configure_internal(win);
+
+    // Set config_id for enabling LWS tuning
+    _config_id = "pooling_layer_3d";
+    _config_id += lower_string(string_from_data_type(data_type));
+    _config_id += "_";
+    _config_id += lower_string(string_from_data_layout(_data_layout));
+    _config_id += "_";
+    _config_id += support::cpp11::to_string(dst->dimension(idx_width));
+    _config_id += "_";
+    _config_id += support::cpp11::to_string(dst->dimension(idx_height));
+    _config_id += "_";
+    _config_id += support::cpp11::to_string(dst->dimension(idx_channel));
+    _config_id += "_";
+    _config_id += lower_string(string_from_data_layout(src->data_layout()));
+
+    ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
+}
+
+Status ClPool3dKernel::validate(const ITensorInfo *src, const ITensorInfo *dst, const Pooling3dLayerInfo &pool_info)
+{
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src, dst, pool_info));
+    return Status{};
+}
+
+void ClPool3dKernel::run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue)
+{
+    ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
+    ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
+
+    const 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_0));
+
+    // Collapse 3D window
+    Window window_collapsed = window.collapse_if_possible(ICLKernel::window(), Window::DimZ);
+
+    // Set CL kernel arguments
+    unsigned int idx = 0;
+    // Passing of the window not needed, as the steps are not used for the pool3d kernel
+    add_5D_tensor_argument(idx, src, window);
+    add_5D_tensor_argument(idx, dst, window);
+    enqueue(queue, *this, window_collapsed, lws_hint());
+}
+} // namespace kernels
+} // namespace opencl
+} // namespace arm_compute
diff --git a/src/gpu/cl/kernels/ClPool3dKernel.h b/src/gpu/cl/kernels/ClPool3dKernel.h
new file mode 100644
index 0000000..30c76ed
--- /dev/null
+++ b/src/gpu/cl/kernels/ClPool3dKernel.h
@@ -0,0 +1,73 @@
+/*
+ * Copyright (c) 2022 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_POOL3D_KERNEL_H
+#define ARM_COMPUTE_CL_POOL3D_KERNEL_H
+
+#include "src/core/common/Macros.h"
+#include "src/gpu/cl/ClCompileContext.h"
+#include "src/gpu/cl/IClKernel.h"
+
+namespace arm_compute
+{
+namespace opencl
+{
+namespace kernels
+{
+/** Interface for the pooling layer kernel */
+class ClPool3dKernel : public IClKernel
+{
+public:
+    ClPool3dKernel();
+    ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClPool3dKernel);
+
+    /** Configure kernel for a given list of arguments
+     *
+     * @note Asymmetric padding is not supported when dimension rounding type == CEIL.
+     *
+     * @param[in]  compile_context The compile context to be used.
+     * @param[in]  src             Source tensor info. Data types supported: F16/F32.
+     * @param[out] dst             Destination tensor info. Data types supported: same as @p src.
+     * @param[in]  pool_info       Contains pooling operation information described in @ref Pooling3dLayerInfo.
+     */
+    void configure(const ClCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst, const Pooling3dLayerInfo &pool_info);
+    /** Static function to check if given info will lead to a valid configuration
+     *
+     * Similar to ClPool3dKernel::configure()
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *src, const ITensorInfo *dst, const Pooling3dLayerInfo &pool_info);
+
+    // Inherited methods overridden:
+    void run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue) override;
+
+private:
+    Pooling3dLayerInfo _pool_info{};
+    DataLayout         _data_layout{ DataLayout::UNKNOWN };
+    unsigned int       _num_elems_processed_per_iteration{ 1 };
+};
+} // namespace kernels
+} // namespace opencl
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_CL_POOL3D_KERNEL_H */
diff --git a/src/gpu/cl/operators/ClPool3d.cpp b/src/gpu/cl/operators/ClPool3d.cpp
new file mode 100644
index 0000000..7dec6c5
--- /dev/null
+++ b/src/gpu/cl/operators/ClPool3d.cpp
@@ -0,0 +1,57 @@
+/*
+ * Copyright (c) 2022 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/gpu/cl/operators/ClPool3d.h"
+
+#include "arm_compute/runtime/CL/CLScheduler.h"
+
+#include "src/gpu/cl/ClCompileContext.h"
+#include "src/gpu/cl/kernels/ClPool3dKernel.h"
+
+#include "src/common/utils/Log.h"
+
+namespace arm_compute
+{
+namespace opencl
+{
+void ClPool3d::configure(const ClCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst, const Pooling3dLayerInfo &info)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(src);
+    ARM_COMPUTE_LOG_PARAMS(src, dst, info);
+
+    // Configure pooling kernel
+    auto k = std::make_unique<kernels::ClPool3dKernel>();
+    k->set_target(CLScheduler::get().target());
+    k->configure(compile_context, src, dst, info);
+    _kernel = std::move(k);
+
+    // Tune kernels
+    CLScheduler::get().tune_kernel_static(*_kernel);
+}
+
+Status ClPool3d::validate(const ITensorInfo *src, const ITensorInfo *dst, const Pooling3dLayerInfo &info)
+{
+    return kernels::ClPool3dKernel::validate(src, dst, info);
+}
+} // namespace opencl
+} // namespace arm_compute
diff --git a/src/gpu/cl/operators/ClPool3d.h b/src/gpu/cl/operators/ClPool3d.h
new file mode 100644
index 0000000..7d994fd
--- /dev/null
+++ b/src/gpu/cl/operators/ClPool3d.h
@@ -0,0 +1,65 @@
+/*
+ * Copyright (c) 2022 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_POOL3D_H
+#define ARM_COMPUTE_CL_POOL3D_H
+
+#include "src/gpu/cl/ClCompileContext.h"
+#include "src/gpu/cl/IClOperator.h"
+
+#include <memory>
+
+namespace arm_compute
+{
+namespace opencl
+{
+/** Basic function to simulate a pooling layer with the specified pooling operation. This function calls the following OpenCL kernels:
+ *
+ * -# @ref opencl::ClPool3d
+ */
+class ClPool3d : public IClOperator
+{
+public:
+    /** Constructor */
+    ClPool3d() = default;
+    /** Configure operator for a given list of arguments
+     *
+     * @note Asymmetric padding is not supported when dimension rounding type == CEIL.
+     *
+     * @param[in]  compile_context The compile context to be used.
+     * @param[in]  src             Source tensor info.
+     * @param[out] dst             Destination tensor info.
+     * @param[in]  info            3d Pooling layer parameters.
+     */
+    void configure(const ClCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst, const Pooling3dLayerInfo &info);
+    /** Static function to check if given info will lead to a valid configuration
+     *
+     * Similar to ClPool3d::configure()
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *src, const ITensorInfo *dst, const Pooling3dLayerInfo &info);
+};
+} // namespace opencl
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_CL_POOL3D_H */
diff --git a/src/runtime/CL/functions/CLPooling3dLayer.cpp b/src/runtime/CL/functions/CLPooling3dLayer.cpp
new file mode 100644
index 0000000..11ae1d0
--- /dev/null
+++ b/src/runtime/CL/functions/CLPooling3dLayer.cpp
@@ -0,0 +1,73 @@
+/*
+ * Copyright (c) 2022 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 "arm_compute/runtime/CL/functions/CLPooling3dLayer.h"
+
+#include "arm_compute/core/CL/CLKernelLibrary.h"
+#include "arm_compute/core/CL/ICLTensor.h"
+#include "src/core/CL/ICLKernel.h"
+#include "src/gpu/cl/operators/ClPool3d.h"
+
+namespace arm_compute
+{
+struct CLPooling3dLayer::Impl
+{
+    const ICLTensor                  *src{ nullptr };
+    ICLTensor                        *dst{ nullptr };
+    ICLTensor                        *indices{ nullptr };
+    std::unique_ptr<opencl::ClPool3d> op{ nullptr };
+};
+
+CLPooling3dLayer::CLPooling3dLayer()
+    : _impl(std::make_unique<Impl>())
+{
+}
+CLPooling3dLayer::~CLPooling3dLayer() = default;
+
+void CLPooling3dLayer::configure(const ICLTensor *input, ICLTensor *output, const Pooling3dLayerInfo &pool_info)
+{
+    configure(CLKernelLibrary::get().get_compile_context(), input, output, pool_info);
+}
+
+void CLPooling3dLayer::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, const Pooling3dLayerInfo &pool_info)
+{
+    _impl->src = input;
+    _impl->dst = output;
+
+    _impl->op = std::make_unique<opencl::ClPool3d>();
+    _impl->op->configure(compile_context, input->info(), output->info(), pool_info);
+}
+
+Status CLPooling3dLayer::validate(const ITensorInfo *input, const ITensorInfo *output, const Pooling3dLayerInfo &pool_info)
+{
+    return opencl::ClPool3d::validate(input, output, pool_info);
+}
+
+void CLPooling3dLayer::run()
+{
+    ITensorPack pack;
+    pack.add_tensor(TensorType::ACL_SRC, _impl->src);
+    pack.add_tensor(TensorType::ACL_DST_0, _impl->dst);
+    _impl->op->run(pack);
+}
+} // namespace arm_compute