There are many fundamental differences between PDF and Word making the approach you want highly undesirable as a general workflow. I'll give just one example: PDF typically does not store information about document structure - sentences, paragraphs, columns, tables... All it stores is the actual text at certain locations at a page. Word of course does have those concepts.
Is it possible to do what you want? Yes, to some extent. In the general case with guesswork and approximation. If you know which information you want to convert it might be possible to search for it in the PDF file generated by SSRS and then generate a Word file out of it. However, if SSRS allows export to text, XML, RTF or any other structure based file format (however slightly structure based), you'd have a much easier time.
If you insist on doing what you suggest here, you would have to:
1) Write code to take the PDF exported from SSRS and interpret it (find the textual content you want)
2) Recreate the necessary structural information from that information (what are paragraphs, where and what are the tables, what's the formatting etc...)
3) Write that into a file Word can read (or create a new Word document directly using automation).
This would be a considerable amount of work, but you have all of the necessary information as the PDF specification is freely downloadable from the Adobe web site and it contains all of the information you need.