Sidestepping Deployment Bottlenecks Before they Happen: tinyML Models Using Virtual Targets