As the other commenters have said, it would be much better to refactor the database to a single table. If you can't do that then the only other thing that I can think of is to have a class which will select the table for you.
I would create a new class which has the same properties as your company tables, and also has the descriminator property. This would then be used as the data source for your ui.
in this class you would have to code manually to draw the data from the correct actual table (and save to it) based on the value of the discriminator. This is fine if you have only a few tables, but as your number of identical tables grows large, this will become more of a headache.
It might be possible to have the base tables all inherit from a virtual base class which would help a bit - you could then create a dictionary which the base class could use to switch the final data source on the fly.
As a final thought have you considered:
1. Creating the master table as suggested by the other commentators as a single table and then having views for each company.
- Creating the master table as suggested and then having code to create the individual tables from that one at some point prior to their use?