You could use a struct with two bytes, manually pack/unpack (dividing by 2^15) and you have the equivalent of one component of FVector_NetQuantizeNormal
FVector_NetQuantizeNormal can be used for more than just -1 to 1 if you multiply by a scalar (I just wrote a motion controller replication thing using it multiplied by 1000 to get a range from -10m to 10m with sub millimeter accuracy).