Are there any data limitations to perform multi-class prediction models?

Data should be pre-processed i.e., encoded & checked for missing values. The target variable should have limited number of unique values. (Unlike regression which have continuous values).