Target Layer Regularization for Continual Learning Using Cramer-Wold Generator