Thank you for quick reply Dr.Karim.
Usually when we use flows from any hydrological model for flood prediction we don't want to miss the peak flows while calibrating the model on a daily basis and also on the other hand we don't want to miss the threshold of NS value (0.6 or 0.7). So, Normalized RMSE = RMSE/mean(Q-observed) is also taken as a constraint along with our defined objective function. Since, NS is estimated using daily observed and simulated flows It may be misleading sometimes. So Normalized RMSE is also considered as a pseudo-objective function in many scenarios.
Regards,
Gokul