Hi Amit! There are some fantastic answers to questions similar to yours on GIS StackExchange, utilising both route and raster based approaches. Have a look at the below and see if it helps!

1) http://gis.stackexchange.com/questions/8685/workflow-for-determining-stream-gradient

2) http://gis.stackexchange.com/questions/9532/stream-gradient-assigning-points-to-their-respective-reaches

3) http://gis.stackexchange.com/questions/11326/stream-gradient-two-types-using-nhd-high-resolution-data

4) http://forums.esri.com/Thread.asp?c=93&f=995&t=219923

Hi Amit. This is an interesting question, but it's not just a GIS methodology question. It depends on what precise use you want to put the derived slope values to. One key decision to make when extracting slopes from a DEM is the length of stream channel (L) over which slope is integrated. Selection of a value for this L variable strongly controls the slope distribution obtained, especially the maximum stream or floodplain slopes: the shorter L is, the more likely you are to capture reaches of steep slope. Longer L values, however, tend to smooth out slopes in the distance integration process.

We touch on this issue in this 2009 paper on stream power and slope extraction for DEMs below:

Barker, D.M., Lawler, D.M., Knight, D.W., Morris, D.G., Davies, H.N., and Stewart, E.J. (2009) 'Longitudinal distributions of river flood power: the Combined Automated Flood, Elevation and Stream power (CAFES) methodology'. Earth Surface Processes and Landforms 34, 280-290