Dear Jose,
There are a number (quite a few more than MrBayes) of 16x16 stem RNA models implemented in HyPhy (available via StandardAnalyses>Basic Analyses>AnalyzeDiNucData.bf).
When you say you have only 8 states, I presume you mean that only 8 pairings (AU,UA,CG,GC,GU,UG and 2 more most likely) are actually observed in the alignment?
You could implement your analysis following the lines of, for example the file Examples/SimpleAnalyses/HKY85shared.bf, with a few tricks.
1). You need to create a doublet filter and explicitly exclude the 8 states you don't want to model. For example
Code:DataSetFilter RNASomeStates = CreateFilter (myData,2,"","","AC,AG,UC");
would create a 13 state filter (16 states minus "AC,AG,UC"). This construct also assumes that your data are linearly arranged, i.e. the site 1 is paired with site 2,
site 3 with site 4 etc
2). When defining the rate matrix to model the evolution of your 8 states, it is important to understand how they are ordered (from 1st to last). For 16 states, the arrangement is lexicographic (AA,AC,AG,AU,....,UA,UC,UG,UU). For a reduced set, the ordering will be the same, except the excluded states won't be there. This mapping will help you define the rate matrix properly.
HTH,
Sergei