In order to use CVX, you need to formulate a convex optimization problem. That means, among other things (constraints), your objective function needs to be convex.
{convex} + {convex} is convex. {convex} - {convex} may or may not be convex, so CVX rejects it