I don't think it is possible to do NGS on mRNA and miRNA using the same kit.  However, the kit preps from both could be mixed on the same sequencing lane with appropriate barcoding.  If you wanted you could do lncRNA and mRNA in the same preparation, but you would probably have to resort to ribodepletion to decrease ribosomal RNA and could not use polyA selection. 

We routinely use the Illumina kit for strand-specific sequence.  We find that knowing the strand-specificity can be very helpful in sorting out specific issues.  It is especially invaluable for lncRNA. I generally like 25-50 million reads for polyA mRNA, although this can vary depending on various factors (i.e. on the higher end for ribodepleted).  I think there is a lot more to be learned from NGS than from chips and am a big believer.  However, don't forget that for any experiment that is important, biological replicates should also be run although might work with lower numbers of reads.  With 15 samples I would probably put five samples per lane of a HiSEQ run.

Now it's very popular people turn to RNAseq, cause you suppose to get the whole transcriptome, when Array contains only annotated targets.

Prescott is right, if you plan to do miRNA and other RNAs the same time, you need to be careful on the method of generate library and how to run it. If you only need to start from GE, then you can even try 3' RNAseq, which give you a coarse view on how overall gene expression is, but without other information like, splicing, etc.

microRNA sequencing is better than microarray for miRNA profiling because:

• Highly accurate base-by-base sequencing eliminates cross-hybridization
• Any microRNA can be quantified by millions of short reads
• Does not depend on prior sequence information. Allows new microRNA discovery.
• Single base resolution detects isomiR sequences
• Discriminates mature microRNA from pre-microRNA
• More cost-effective then microRNA microarray