Can we trust elemental abundances derived in late-type giants with the classical 1D stellar atmosphere models?