Text this: A framework for robust deep learning models against adversarial attacks based on a protection layer approach