Sure, easily. On example is vision: you'd just have to adjust input preprocessing in existing models.