Perhaps CQL is not your best choice. CQL requires you to define your schema up front to allow for SQL-like syntax capability, but it takes away some of the useful dynamic features of Cassandra. Using the Thrift-based mutation capability of nearly any Cassandra client, you can simply name the column you want to write, and if it doesn't exist it will be created for you.
For example:
ks.insert(key, createColumnPath(colName), bytes(value));