In the recent OpenCL backend, the kernel is enqueued with:

result = clEnqueueNDRangeKernel(app->configuration.commandQueue[0], axis->kernel, 3, 0, global_work_size, local_work_size, 0, 0, 0);

i.e. the last three arguments are set to 0. It would be nice to set/have access to these event arguments, which are helpful for synchronization and profiling.