Reformulations for utilizing separability when solving convex MINLP problems