RNA (and DNA) entities are very rarely approved as medicines, because one can't usually deliver sufficient quantities of these large molecules to enough cells in the tissues of an existing organism. It's often a game of guessing in which tissues and diseases you can get away with minimal delivery efficiency. Recent notable companies who did this well enough are Isis Pharmaceuticals and uniQure.
If you could generally deliver enough construct, then there might be additional theoretical safety and efficacy questions like the ones you allude to. But that's hypothetical. The hard problem is large molecule delivery.
For protein production, we generally prefer DNA constructs over mRNA, because it's permanent, heritable and thus very easy and inexpensive. If you need a switch (which you often don't), then there are DNA-based switches like the IPTG and lactose autoinduction systems in E.coli, and the tetracycline and steroid-inducible systems in mammalian cells. And/or you could encode an RNA based switch in your DNA construct ("riboswitch").
There is a role for tRNA in protein expression, for purposes of rare codon supplementation. But there too, I prefer to encode the necessary features in DNA, or in modern days you might as well do away with the problem by having your genes synthesized from scratch, which allows you to take full control of codon choices.
