diff --git a/ggml-sycl.cpp b/ggml-sycl.cpp
index 51acebf19..fc4d2964c 100644
--- a/ggml-sycl.cpp
+++ b/ggml-sycl.cpp
@@ -12921,7 +12921,7 @@ int get_work_group_size(int user_device_id) {
     return prop.get_max_work_group_size();
 }
 
-void ggml_init_sycl() try {
+static void ggml_init_sycl() try {
     static bool initialized = false;
 
     if (!initialized) {
@@ -16417,6 +16417,7 @@ static ggml_backend_buffer_type_i ggml_backend_sycl_buffer_type_interface = {
 };
 
 ggml_backend_buffer_type_t ggml_backend_sycl_buffer_type(int device_index) {
+    ggml_init_sycl();
     if (device_index>=g_device_count or device_index<0) {
         printf("ggml_backend_sycl_buffer_type error: device_index:%d is out of range [0, %d], miss to call ggml_backend_sycl_set_single_device()\n",
             device_index, g_device_count-1);
@@ -16786,6 +16787,7 @@ static ggml_backend_buffer_type_i ggml_backend_sycl_split_buffer_type_interface
 };
 
 GGML_CALL ggml_backend_buffer_type_t ggml_backend_sycl_split_buffer_type(const float * tensor_split) {
+    ggml_init_sycl();
     // FIXME: this is not thread safe
     static std::map<std::array<float, GGML_SYCL_MAX_DEVICES>, struct ggml_backend_buffer_type> buft_map;
 
@@ -17153,7 +17155,7 @@ static ggml_guid_t ggml_backend_sycl_guid() {
 }
 
 GGML_CALL ggml_backend_t ggml_backend_sycl_init(int device) {
-    ggml_init_sycl(); // TODO: remove from ggml.c
+    ggml_init_sycl();
 
     check_allow_gpu_index(device);
 
diff --git a/ggml-sycl.h b/ggml-sycl.h
index 00c82f7b8..a9f776fc1 100644
--- a/ggml-sycl.h
+++ b/ggml-sycl.h
@@ -16,9 +16,6 @@ extern "C" {
 #define GGML_SYCL_MAX_DEVICES       48
 #define GGML_SYCL_NAME "SYCL"
 
-GGML_API void   ggml_init_sycl(void);
-GGML_API bool   ggml_sycl_compute_forward(struct ggml_compute_params * params, struct ggml_tensor * tensor);
-
 // backend API
 GGML_API ggml_backend_t ggml_backend_sycl_init(int device);
 
diff --git a/ggml.c b/ggml.c
index d1a2775f2..9017637ff 100644
--- a/ggml.c
+++ b/ggml.c
@@ -286,8 +286,6 @@ inline static void * ggml_calloc(size_t num, size_t size) {
 #include "ggml-opencl.h"
 #elif defined(GGML_USE_VULKAN)
 #include "ggml-vulkan.h"
-#elif defined(GGML_USE_SYCL)
-#include "ggml-sycl.h"
 #endif
 
 // floating point type used to accumulate sums
@@ -2642,8 +2640,6 @@ struct ggml_context * ggml_init(struct ggml_init_params params) {
         ggml_cl_init();
 #elif defined(GGML_USE_VULKAN)
         ggml_vk_init_cpu_assist();
-#elif defined(GGML_USE_SYCL)
-        ggml_init_sycl();
 #endif
 
         ggml_setup_op_has_task_pass();