A coarse protocol idea would be to run your extracts on a gel or size exclusion columns, while making sure you work in RNase free conditions, reverse transcribe the result, ligate adaptors in 3’ and 5’ - or first ligate and then RT which is a bit more tricky - and amplify the product for subsequent analysis. Alternatively, you could look into RACE (rapid amplification of cDNA ends), some protocols are specifically adapted for sRNA detection, eg. 5’tagRACE.