Efficient eval of Linear-Model

I'm using the statistics toolbox, specifically stepwiselm. Everything works fine.

My question is this: is there a function with which to evaluate the model that doesn't carry around the overhead of the LinearModel object. For example, is there a function that behaves like:

% create the model
mdl = stepwiselm(X,y,'poly45')

% evaluate the model just using Terms and Coefficients at new values of X
yhat = modelValue( mdl.Formula.Terms, mdl.Coefficients.Estimate, Xnew);

The idea is that Terms and Coefficients arrays are small (1K bytes) and I can store them for later use, but the model (mdl) is large (500K bytes), containing all the data and other stuff. (I'm gonna have "a lot" of models.)

Thanks.