Hi Shantanu,
In FSL you can run the method of Inter-Repetition Variance Analysis (IRVA) on an event-related task involving several repeated stimuli. Without assuming anything about the shape of the hemodynamic response function, you run ANOVA on the repeated time sequences of the signal intensity, and test the significance at a voxel by comparing the within variance with the between variance.
In terms of no assumption regarding the shape of hrf, AFNI does have such a similar approach by running 3dDeconvolve in its deconvolution mode (instead of its regression mode with outputs from waver), but the underlying mechanism is different in the sense that AFNI still tries to model the hrf in the process. I am not so sure which approach would work better under various circumstances.
Gang