Data IO :
I think every functions are existed in armadillo/mlpack already,
except of SQL import/export. SQL import/export need to rely introduce
new dependencies, may not a good idea for mlpack, I think this feature
could be skipped.

Data Transformation :
Filter : Now we have MissingPolicy, in the future we could add more policies if needed
Manipulation :
Split--implemented
Join--armadillo provided
Convert numerical<-->categorical, done already
Normalization--armadillo provide lots of functions already
Sampling--haven't done yet, I do not know what kind of api we should provide
for Sampling, I think every dataset have their own needs
Clean missing data--done by imputation classes

Statistical Analysis :
Most of the functions are provided by armadillo.I think provide statistic functions
like R language could be quite useful too(like summary and describe functions).

Mathematical Operator :
I think most functions are provided by armadillo already, except
of Time Zone, not quite sure what Keon what to do with this.