Do you want to then run statistical tests on the IRF's? Whether you can do that or not for your blocked design depends on how your experiment is set up: e.g., deconvolution requires more datapoints than simple regression using waver; stimulus timing should be jittered or spaced so that the reponses can be teased apart. Or do you want to estimate IRF's simply as a convenient method for examining timeseries data, and run 3dDeconvolve separately with convolved waveforms to calculate your statistics?