What you want to do is generally referred to as 'cellular deconvolution'.
As far as I am aware, there is no 'leading' method yet available for RNA-seq, yet alone scRNA-seq; however, some have used CIBERSORT, which was originally developed for microarray data. Another program that was developed for microarray data was called CellMix, which mainly used a gene signature called 'Abbas' - available in R but not suitable for RNA-seq.
I believe people generally manually infer cell-type after they perform some transformation on their data, such as tSNE. With tSNE, you can check the expression of each gene in each identified cluster and, through this, identify the key genes relating to each cluster. One can then infer the likely cell-type.
Another simple way to do it is to transform your data into Z-scores and pick genes that have absolute Z-score > 2 or 3 in each cell. By doing this, you can identify the most highly and lowly expressed genes in each cluster, and thus attempt to infer the cell type.
I also found another way to identify key genes in each cluster, but on CyTOF data: cytofNet (see steps 8 and 9).
Kevin