If I am reading correctly, you have a pretty straightforward design, which could be analyzed in a multiple regression/general linear model (GLM) framework, using both the categorical variables (type of dance and skill level) as well as the 'years training' continuous variable as predictors. One way of handling the multiple DVs would be to just run the model separately for each outcome (its not the "best" analysis/idea but it would work, especially if you use an alpha adjustment over all the tests you do).
Each analysis answers a different question - make sure you clearly phrase your research questions - all of them - explicitly. This always guides analysis. For example, you may want to know "Does the type of dance predict body image (measured using test C) all other variables being constant?, " vs "Does type of dance interact with skill level to predict body image (measured using test A?)"
Write out each of your potential questions as clearly as possible, exhausting all the questions you want to answer. Then think about which analyses will answer them, one by one. Use an alpha adjustment to control for multiple tests.