Electromagnetic waveforms are an essential component of high-fidelity radar and electronic warfare digital computer simulations. Sampled representations of radar waveforms are widely used for their physical realism and suitability for algorithimic processing. However, this fidelity comes at a price because operations on radar waveforms are often a computationally costly simulation bottleneck. In this report, we propose a method for constructing a reduced, feature-based alternative radar waveform model component derived from a given high-fidelity component. The resulting model will be related to the original through an approximate morphism. The proposed method is illustrated with a highly simplified waveform model. Both linear and nonlinear approaches are considered; in particular, a role for machine learning techniques is identified.