On the Acceleration of a POD-based Model Reduction Technique