Trust-Region Methods with Low-Fidelity Objective Models