For analysis, you’d most likely want a single common set of response functions. If you were using multiple subject- or group-specific responses, you’d need to take the differences between these responses into account when interpreting differences in the ODFs which is not trivial nor recommended.
We typically recommend the group-average as the most representative set of response functions unless you have good reasons to believe that the group-average response would not be representative of a tissue type. If in doubt, I’d take the average across the control subjects so that the interpretation of the responses (and derived ODFs) are tied to normal tissue properties. But note that even when there are subtle differences in the responses between groups, using the group average response is not necessarily a bad strategy as the mismatch between signal and average response could still manifest in group differences in the estimated ODFs (the constraints in CSD allowing). There are a few threads discussing this in more detail if you are interested: