You certainly can map them like the paired-end reads. The question becomes if it's worth your time to bother. If you don't have a lot of orphans like this, then you can just ignore them. If, however, you have a LOT of orphans (e.g., there was a problem with read #2) then you pretty much have to align the orphaned reads. Use the same settings as for the paired-end datasets (within reason).
Edit: I should add that the SE reads might map slightly differently than the PE reads. In many cases this isn't a problem. However if whatever you're doing is very sensitive to any bias then either compare the results with/without including the SE reads or at least ensure that the proportion of SE reads is approximately constant across samples (i.e., try to ensure that all of the samples have a similar bias).