Currently the tokenizer _timelex is a private interface. I think it is worthwhile to make that public and pass the class to parserinfo so that applications built on dateutil can subclass the tokenizer either to optimize its performance for specific use cases or to add compatibility that might not be appropriate in the general case.

We may want to make a similar consideration about _tzparser as well.

I have also mentioned this in ms-boom/dateutil#1 and #122.