Investigate poor performances of TVM

[diegofiori]

The current way the TVM configuration is done, it seems not to exploit the full potential of the TVM library since the performance is lower than theoretically expected (for TVM optimization only).