Part of the problem is that there are realtime systems where, if you don't meet the deadline, the plane will fall out of the sky. Then there are realtime systems like VoIP where, if you don't meet the deadline, the call will cut out.
And there is a desire to get "almost realtime" for things like VoIP on general purpose time sharing systems, where you want to deprioritize less latency sensitive tasks in order to hit the deadline if you can, but nobody dies if you don't.
The trouble with using "real" realtime in those systems is that the system may not have enough capacity in theory even when it does in practice. In other words, if you have three concurrent realtime processes that could each in theory use up to 50% of the available resources at a time, but in practice almost never use more than 10% each, then for less critical systems you may be willing to bank on "almost never" not happening often enough to matter.
There is a wikipedia page for Victor's stuff: https://en.wikipedia.org/wiki/RTLinux
If you want a time sharing system and real time, that problem has been solved, in a really elegant way. Victor Yodaiken did it years ago, he made a small real time kernel and he ran all of Linux as the idle process in the real time kernel. When Linux thought it was disabling interrupts it wasn't, only the real time kernel could do that.
He had a demo where he was running xperf (or some fairly compute intensive X thing) while running
tar cf - / | rsh someotherhost 'cat > /dev/null'
while the real time kernel was gathering events. He never missed a deadline or dropped an event.
Sweetest design I've ever seen, have your cake and eat it too.