Precompile GPU ML models.
Prepares TensorFlow features and applies known restrictions on some devices.