There are three distinct parts to this:
* Getting access to samples - should be simple via your favourite torrent site. * Writing a decent engine. Tricky, but not impossible if you've got access to the low-level APIs and the technical skills. * Getting reviewed/taken seriously.
The sad truth is few people will pay for a new antivirus tool, especially one with no proven provenance, support team, and real world accuracy. (i.e. You can block all known samples easily, but fail on new viruses/trojans.)