The operation in subthreshold was misunderstood (and ignored) for some time, but now it is well characterized. So, as long as the models you are using make a good job, there is no problem in working there.
Note also that there is nothing "mysterious" about the operation in subthreshold region: the transistor doesn't "know" that it is operating there and starts to make crazy things (also one transistor doesn't "know" that its neighbor is operating in another region, so you can mix them).
The only problem that happened in the past, was that there were no acceptable models for that region.
Rgds
PS: The regions of operation in a transistor are only arbitrary denominations we use. The transistor couldn't care less about our denominations.
We make that distinction because we can identify particular operating properties in each region: for example in subthreshold, the current depends exponentially of the VGS, while in strong inversion there is a quadratic dependence.