A flexible maximum-entropy component separation algorithm is presented that accommodates anisotropic noise, incomplete sky-coverage and uncertainties in the spectral parameters of foregrounds. The capabilities of the method are determined by first applying it to simulated spherical microwave data sets emulating the COBE-DMR, COBE-DIRBE and Haslam surveys. Using these simulations we find that is very difficult to determine unambiguously the spectral parameters of the galactic components for this data set due to their high level of noise. Nevertheless, we show that is possible to find a robust CMB reconstruction, especially at the high galactic latitude. The method is then applied to these real data sets to obtain reconstructions of the CMB component and galactic foreground emission over the whole sky. The best reconstructions are found for values of the spectral parameters: T_d=19 K, alpha_d=2, beta_ff=-0.19 and beta_syn=-0.8. The CMB map has been recovered with an estimated statistical error of sim 22 muK on an angular scale of 7 degrees outside the galactic cut whereas the low galactic latitude region presents contamination from the foreground emissions.