Durbin-Watson test is mainly devoted to lag one autocorrelation in the context of regression with time series. There are variants for higher lags but they are rarely used. The autocorrelation function is more convenient then. Both are covered in all texts in econometrics. Lagrange multiplier testing is more general and is covered in most texts in statistics and econometrics.
As far as I know, the KPSS test is more specific. Its purpose is to test stationarity against the presence of a unit root. This is not really a test of autocorrelation.